Jan Peters

Nico Bohlinger, Jan Peters

We propose to turn generalist multi-embodiment value functions into reusable models for robot design. Instead of running a new reinforcement learning co-design loop for each robot, we first train an embodiment-aware policy and value function across many robot designs. After training, the frozen value function is used as a differentiable surrogate to optimize candidate embodiments through value gradients. We evaluate our approach across different robot design settings, from perturbed single robots to held-out robots across morphology classes, with single models trained on up to 50 robots and design spaces of over 1100 continuous embodiment parameters. Beyond optimizing complete embodiments, we show that value gradients can identify performance-limiting design and control parameters, enabling both the optimization and the analysis of new robot designs.

Joe Watson, Chen Song, Oliver Weeger et al. · cambridge

This survey examines the broad suite of methods and models for combining machine learning with physics knowledge for prediction and forecast, with a focus on partial differential equations. These methods have attracted significant interest due to their potential impact on advancing scientific research and industrial practices by improving predictive models with small- or large-scale datasets and expressive predictive models with useful inductive biases. The survey has two parts. The first considers incorporating physics knowledge on an architectural level through objective functions, structured predictive models, and data augmentation. The second considers data as physics knowledge, which motivates looking at multi-task, meta, and contextual learning as an alternative approach to incorporating physics knowledge in a data-driven fashion. Finally, we also provide an industrial perspective on the application of these methods and a survey of the open-source ecosystem for physics-informed machine learning.

Julen Urain, Niklas Funk, Jan Peters et al.

Multi-objective optimization problems are ubiquitous in robotics, e.g., the optimization of a robot manipulation task requires a joint consideration of grasp pose configurations, collisions and joint limits. While some demands can be easily hand-designed, e.g., the smoothness of a trajectory, several task-specific objectives need to be learned from data. This work introduces a method for learning data-driven SE(3) cost functions as diffusion models. Diffusion models can represent highly-expressive multimodal distributions and exhibit proper gradients over the entire space due to their score-matching training objective. Learning costs as diffusion models allows their seamless integration with other costs into a single differentiable objective function, enabling joint gradient-based motion optimization. In this work, we focus on learning SE(3) diffusion models for 6DoF grasping, giving rise to a novel framework for joint grasp and motion optimization without needing to decouple grasp selection from trajectory generation. We evaluate the representation power of our SE(3) diffusion models w.r.t. classical generative models, and we showcase the superior performance of our proposed optimization framework in a series of simulated and real-world robotic manipulation tasks against representative baselines.

Christian Scherer, Joe Watson, Theo Gruner et al.

Distilling expert demonstration data into large generative models using behavioral cloning is a scalable approach to learning capable policies for robotic control, particularly for dexterous manipulation. Reinforcement learning (RL) can be used as a means to finetune these policies further using additional experience. An open question is whether RL is more sample-efficient than collecting more human demonstrations. Prior work has finetuned large pretrained policies in a scalable fashion by applying RL to a smaller residual policy that corrects the pretrained model. However, for the typical sparse reward tasks, RL algorithms can struggle to optimize the behavior in a sample-efficient manner. We explore inverse reinforcement learning, where a dense reward function is learned from expert demonstrations, potentially reducing the challenge of RL finetuning. We specifically consider coherent imitation learning, an IRL method that facilitates improvement of the BC policy through using a specific reward formulation with theoretical guarantees. We show that our IRL method maintains or improves the performance of pi-0.5 on all six sparse manipulation tasks and achieves a $\geq 90\%$ success rate on five out of six complex manipulation tasks, outperforming RL-based baselines using sparse rewards. By ensuring our initial pretrained finetuning policy is optimal for our initial reward and critic, our method circumvents the initial drop commonly seen in RL finetuning and enables faster improvement.

Snehal Jauhri, Jan Peters, Georgia Chalvatzaki

Mobile Manipulation (MM) systems are ideal candidates for taking up the role of a personal assistant in unstructured real-world environments. Among other challenges, MM requires effective coordination of the robot's embodiments for executing tasks that require both mobility and manipulation. Reinforcement Learning (RL) holds the promise of endowing robots with adaptive behaviors, but most methods require prohibitively large amounts of data for learning a useful control policy. In this work, we study the integration of robotic reachability priors in actor-critic RL methods for accelerating the learning of MM for reaching and fetching tasks. Namely, we consider the problem of optimal base placement and the subsequent decision of whether to activate the arm for reaching a 6D target. For this, we devise a novel Hybrid RL method that handles discrete and continuous actions jointly, resorting to the Gumbel-Softmax reparameterization. Next, we train a reachability prior using data from the operational robot workspace, inspired by classical methods. Subsequently, we derive Boosted Hybrid RL (BHyRL), a novel algorithm for learning Q-functions by modeling them as a sum of residual approximators. Every time a new task needs to be learned, we can transfer our learned residuals and learn the component of the Q-function that is task-specific, hence, maintaining the task structure from prior behaviors. Moreover, we find that regularizing the target policy with a prior policy yields more expressive behaviors. We evaluate our method in simulation in reaching and fetching tasks of increasing difficulty, and we show the superior performance of BHyRL against baseline methods. Finally, we zero-transfer our learned 6D fetching policy with BHyRL to our MM robot TIAGo++. For more details and code release, please refer to our project site: irosalab.com/rlmmbp

Puze Liu, Kuo Zhang, Davide Tateo et al.

Autonomous robots should operate in real-world dynamic environments and collaborate with humans in tight spaces. A key component for allowing robots to leave structured lab and manufacturing settings is their ability to evaluate online and real-time collisions with the world around them. Distance-based constraints are fundamental for enabling robots to plan their actions and act safely, protecting both humans and their hardware. However, different applications require different distance resolutions, leading to various heuristic approaches for measuring distance fields w.r.t. obstacles, which are computationally expensive and hinder their application in dynamic obstacle avoidance use-cases. We propose Regularized Deep Signed Distance Fields (ReDSDF), a single neural implicit function that can compute smooth distance fields at any scale, with fine-grained resolution over high-dimensional manifolds and articulated bodies like humans, thanks to our effective data generation and a simple inductive bias during training. We demonstrate the effectiveness of our approach in representative simulated tasks for whole-body control (WBC) and safe Human-Robot Interaction (HRI) in shared workspaces. Finally, we provide proof of concept of a real-world application in a HRI handover task with a mobile manipulator robot.

Carlos E. Luis, Alessandro G. Bottero, Julia Vinogradska et al.

Quantifying uncertainty about a policy's long-term performance is important to solve sequential decision-making tasks. We study the problem from a model-based Bayesian reinforcement learning perspective, where the goal is to learn the posterior distribution over value functions induced by parameter (epistemic) uncertainty of the Markov decision process. Previous work restricts the analysis to a few moments of the distribution over values or imposes a particular distribution shape, e.g., Gaussians. Inspired by distributional reinforcement learning, we introduce a Bellman operator whose fixed-point is the value distribution function. Based on our theory, we propose Epistemic Quantile-Regression (EQR), a model-based algorithm that learns a value distribution function. We combine EQR with soft actor-critic (SAC) for policy optimization with an arbitrary differentiable objective function of the learned value distribution. Evaluation across several continuous-control tasks shows performance benefits with respect to both model-based and model-free algorithms. The code is available at https://github.com/boschresearch/dist-mbrl.

Bang You, Oleg Arenz, Youping Chen et al.

Recent methods for reinforcement learning from images use auxiliary tasks to learn image features that are used by the agent's policy or Q-function. In particular, methods based on contrastive learning that induce linearity of the latent dynamics or invariance to data augmentation have been shown to greatly improve the sample efficiency of the reinforcement learning algorithm and the generalizability of the learned embedding. We further argue, that explicitly improving Markovianity of the learned embedding is desirable and propose a self-supervised representation learning method which integrates contrastive learning with dynamic models to synergistically combine these three objectives: (1) We maximize the InfoNCE bound on the mutual information between the state- and action-embedding and the embedding of the next state to induce a linearly predictive embedding without explicitly learning a linear transition model, (2) we further improve Markovianity of the learned embedding by explicitly learning a non-linear transition model using regression, and (3) we maximize the mutual information between the two nonlinear predictions of the next embeddings based on the current action and two independent augmentations of the current state, which naturally induces transformation invariance not only for the state embedding, but also for the nonlinear transition model. Experimental evaluation on the Deepmind control suite shows that our proposed method achieves higher sample efficiency and better generalization than state-of-art methods based on contrastive learning or reconstruction.

Piotr Kicki, Puze Liu, Davide Tateo et al.

Motion planning is a mature area of research in robotics with many well-established methods based on optimization or sampling the state space, suitable for solving kinematic motion planning. However, when dynamic motions under constraints are needed and computation time is limited, fast kinodynamic planning on the constraint manifold is indispensable. In recent years, learning-based solutions have become alternatives to classical approaches, but they still lack comprehensive handling of complex constraints, such as planning on a lower-dimensional manifold of the task space while considering the robot's dynamics. This paper introduces a novel learning-to-plan framework that exploits the concept of constraint manifold, including dynamics, and neural planning methods. Our approach generates plans satisfying an arbitrary set of constraints and computes them in a short constant time, namely the inference time of a neural network. This allows the robot to plan and replan reactively, making our approach suitable for dynamic environments. We validate our approach on two simulated tasks and in a demanding real-world scenario, where we use a Kuka LBR Iiwa 14 robotic arm to perform the hitting movement in robotic Air Hockey.

Shangding Gu, Alap Kshirsagar, Yali Du et al.

Deployment of Reinforcement Learning (RL) algorithms for robotics applications in the real world requires ensuring the safety of the robot and its environment. Safe Robot RL (SRRL) is a crucial step towards achieving human-robot coexistence. In this paper, we envision a human-centered SRRL framework consisting of three stages: safe exploration, safety value alignment, and safe collaboration. We examine the research gaps in these areas and propose to leverage interactive behaviors for SRRL. Interactive behaviors enable bi-directional information transfer between humans and robots, such as conversational robot ChatGPT. We argue that interactive behaviors need further attention from the SRRL community. We discuss four open challenges related to the robustness, efficiency, transparency, and adaptability of SRRL with interactive behaviors.

Joe Watson, Jan Peters

Monte Carlo methods have become increasingly relevant for control of non-differentiable systems, approximate dynamics models and learning from data. These methods scale to high-dimensional spaces and are effective at the non-convex optimizations often seen in robot learning. We look at sample-based methods from the perspective of inference-based control, specifically posterior policy iteration. From this perspective, we highlight how Gaussian noise priors produce rough control actions that are unsuitable for physical robot deployment. Considering smoother Gaussian process priors, as used in episodic reinforcement learning and motion planning, we demonstrate how smoother model predictive control can be achieved using online sequential inference. This inference is realized through an efficient factorization of the action distribution and a novel means of optimizing the likelihood temperature to improve importance sampling accuracy. We evaluate this approach on several high-dimensional robot control tasks, matching the sample efficiency of prior heuristic methods while also ensuring smoothness. Simulation results can be seen at https://monte-carlo-ppi.github.io/.

Puze Liu, Kuo Zhang, Davide Tateo et al.

Safety is a crucial property of every robotic platform: any control policy should always comply with actuator limits and avoid collisions with the environment and humans. In reinforcement learning, safety is even more fundamental for exploring an environment without causing any damage. While there are many proposed solutions to the safe exploration problem, only a few of them can deal with the complexity of the real world. This paper introduces a new formulation of safe exploration for reinforcement learning of various robotic tasks. Our approach applies to a wide class of robotic platforms and enforces safety even under complex collision constraints learned from data by exploring the tangent space of the constraint manifold. Our proposed approach achieves state-of-the-art performance in simulated high-dimensional and dynamic tasks while avoiding collisions with the environment. We show safe real-world deployment of our learned controller on a TIAGo++ robot, achieving remarkable performance in manipulation and human-robot interaction tasks.

Kay Hansel, Julen Urain, Jan Peters et al.

Motion generation in cluttered, dense, and dynamic environments is a central topic in robotics, rendered as a multi-objective decision-making problem. Current approaches trade-off between safety and performance. On the one hand, reactive policies guarantee fast response to environmental changes at the risk of suboptimal behavior. On the other hand, planning-based motion generation provides feasible trajectories, but the high computational cost may limit the control frequency and thus safety. To combine the benefits of reactive policies and planning, we propose a hierarchical motion generation method. Moreover, we adopt probabilistic inference methods to formalize the hierarchical model and stochastic optimization. We realize this approach as a weighted product of stochastic, reactive expert policies, where planning is used to adaptively compute the optimal weights over the task horizon. This stochastic optimization avoids local optima and proposes feasible reactive plans that find paths in cluttered and dense environments. Our extensive experimental study in planar navigation and 6DoF manipulation shows that our proposed hierarchical motion generation method outperforms both myopic reactive controllers and online re-planning methods.

Max Siebenborn, Boris Belousov, Junning Huang et al.

Decision Transformer (DT) is a recently proposed architecture for Reinforcement Learning that frames the decision-making process as an auto-regressive sequence modeling problem and uses a Transformer model to predict the next action in a sequence of states, actions, and rewards. In this paper, we analyze how crucial the Transformer model is in the complete DT architecture on continuous control tasks. Namely, we replace the Transformer by an LSTM model while keeping the other parts unchanged to obtain what we call a Decision LSTM model. We compare it to DT on continuous control tasks, including pendulum swing-up and stabilization, in simulation and on physical hardware. Our experiments show that DT struggles with continuous control problems, such as inverted pendulum and Furuta pendulum stabilization. On the other hand, the proposed Decision LSTM is able to achieve expert-level performance on these tasks, in addition to learning a swing-up controller on the real system. These results suggest that the strength of the Decision Transformer for continuous control tasks may lie in the overall sequential modeling architecture and not in the Transformer per se.

Lei Xu, Tianyu Ren, Georgia Chalvatzaki et al.

As robots play an increasingly important role in the industrial, the expectations about their applications for everyday living tasks are getting higher. Robots need to perform long-horizon tasks that consist of several sub-tasks that need to be accomplished. Task and Motion Planning (TAMP) provides a hierarchical framework to handle the sequential nature of manipulation tasks by interleaving a symbolic task planner that generates a possible action sequence, with a motion planner that checks the kinematic feasibility in the geometric world, generating robot trajectories if several constraints are satisfied, e.g., a collision-free trajectory from one state to another. Hence, the reasoning about the task plan's geometric grounding is taken over by the motion planner. However, motion planning is computationally intense and is usability as feasibility checker casts TAMP methods inapplicable to real-world scenarios. In this paper, we introduce neural feasibility classifier (NFC), a simple yet effective visual heuristic for classifying the feasibility of proposed actions in TAMP. Namely, NFC will identify infeasible actions of the task planner without the need for costly motion planning, hence reducing planning time in multi-step manipulation tasks. NFC encodes the image of the robot's workspace into a feature map thanks to convolutional neural network (CNN). We train NFC using simulated data from TAMP problems and label the instances based on IK feasibility checking. Our empirical results in different simulated manipulation tasks show that our NFC generalizes to the entire robot workspace and has high prediction accuracy even in scenes with multiple obstructions. When combined with state-of-the-art integrated TAMP, our NFC enhances its performance while reducing its planning time.

Niklas Funk, Svenja Menzenbach, Georgia Chalvatzaki et al.

Robot assembly discovery is a challenging problem that lives at the intersection of resource allocation and motion planning. The goal is to combine a predefined set of objects to form something new while considering task execution with the robot-in-the-loop. In this work, we tackle the problem of building arbitrary, predefined target structures entirely from scratch using a set of Tetris-like building blocks and a robotic manipulator. Our novel hierarchical approach aims at efficiently decomposing the overall task into three feasible levels that benefit mutually from each other. On the high level, we run a classical mixed-integer program for global optimization of block-type selection and the blocks' final poses to recreate the desired shape. Its output is then exploited to efficiently guide the exploration of an underlying reinforcement learning (RL) policy. This RL policy draws its generalization properties from a flexible graph-based representation that is learned through Q-learning and can be refined with search. Moreover, it accounts for the necessary conditions of structural stability and robotic feasibility that cannot be effectively reflected in the previous layer. Lastly, a grasp and motion planner transforms the desired assembly commands into robot joint movements. We demonstrate our proposed method's performance on a set of competitive simulated RAD environments, showcase real-world transfer, and report performance and robustness gains compared to an unstructured end-to-end approach. Videos are available at https://sites.google.com/view/rl-meets-milp .

Tim Schneider, Boris Belousov, Georgia Chalvatzaki et al.

Robotic manipulation stands as a largely unsolved problem despite significant advances in robotics and machine learning in recent years. One of the key challenges in manipulation is the exploration of the dynamics of the environment when there is continuous contact between the objects being manipulated. This paper proposes a model-based active exploration approach that enables efficient learning in sparse-reward robotic manipulation tasks. The proposed method estimates an information gain objective using an ensemble of probabilistic models and deploys model predictive control (MPC) to plan actions online that maximize the expected reward while also performing directed exploration. We evaluate our proposed algorithm in simulation and on a real robot, trained from scratch with our method, on a challenging ball pushing task on tilted tables, where the target ball position is not known to the agent a-priori. Our real-world robot experiment serves as a fundamental application of active exploration in model-based reinforcement learning of complex robotic manipulation tasks.

Vignesh Prasad, Dorothea Koert, Ruth Stock-Homburg et al.

Modeling interaction dynamics to generate robot trajectories that enable a robot to adapt and react to a human's actions and intentions is critical for efficient and effective collaborative Human-Robot Interactions (HRI). Learning from Demonstration (LfD) methods from Human-Human Interactions (HHI) have shown promising results, especially when coupled with representation learning techniques. However, such methods for learning HRI either do not scale well to high dimensional data or cannot accurately adapt to changing via-poses of the interacting partner. We propose Multimodal Interactive Latent Dynamics (MILD), a method that couples deep representation learning and probabilistic machine learning to address the problem of two-party physical HRIs. We learn the interaction dynamics from demonstrations, using Hidden Semi-Markov Models (HSMMs) to model the joint distribution of the interacting agents in the latent space of a Variational Autoencoder (VAE). Our experimental evaluations for learning HRI from HHI demonstrations show that MILD effectively captures the multimodality in the latent representations of HRI tasks, allowing us to decode the varying dynamics occurring in such tasks. Compared to related work, MILD generates more accurate trajectories for the controlled agent (robot) when conditioned on the observed agent's (human) trajectory. Notably, MILD can learn directly from camera-based pose estimations to generate trajectories, which we then map to a humanoid robot without the need for any additional training.

Hamish Flynn, David Reeb, Melih Kandemir et al.

We present improved algorithms with worst-case regret guarantees for the stochastic linear bandit problem. The widely used "optimism in the face of uncertainty" principle reduces a stochastic bandit problem to the construction of a confidence sequence for the unknown reward function. The performance of the resulting bandit algorithm depends on the size of the confidence sequence, with smaller confidence sets yielding better empirical performance and stronger regret guarantees. In this work, we use a novel tail bound for adaptive martingale mixtures to construct confidence sequences which are suitable for stochastic bandits. These confidence sequences allow for efficient action selection via convex programming. We prove that a linear bandit algorithm based on our confidence sequences is guaranteed to achieve competitive worst-case regret. We show that our confidence sequences are tighter than competitors, both empirically and theoretically. Finally, we demonstrate that our tighter confidence sequences give improved performance in several hyperparameter tuning tasks.

An T. Le, Kay Hansel, Jan Peters et al.

We present hierarchical policy blending as optimal transport (HiPBOT). HiPBOT hierarchically adjusts the weights of low-level reactive expert policies of different agents by adding a look-ahead planning layer on the parameter space. The high-level planner renders policy blending as unbalanced optimal transport consolidating the scaling of the underlying Riemannian motion policies. As a result, HiPBOT effectively decides the priorities between expert policies and agents, ensuring the task's success and guaranteeing safety. Experimental results in several application scenarios, from low-dimensional navigation to high-dimensional whole-body control, show the efficacy and efficiency of HiPBOT. Our method outperforms state-of-the-art baselines -- either adopting probabilistic inference or defining a tree structure of experts -- paving the way for new applications of optimal transport to robot control. More material at https://sites.google.com/view/hipobot

Tobias Lademann, Théo Vincent, Jan Peters et al.

Reinforcement learning has shown promising results for optimizing the control of industrial energy systems, yet most existing studies remain limited to the application in simulation environments. We investigate the challenges of deploying reinforcement learning in a real-world industrial energy system, considering a thermal heating network as a use case. We formulate the task as a Markov Decision Process and systematically analyze the associated challenges along the structure of the formal description, including partial observability, action space design, reward design, and the simulation-to-reality gap. The challenges are grounded in an existing real-world deployment, where reinforcement learning achieves operational stability but shows a significant performance gap compared to simulation.

Julen Urain, An T. Le, Alexander Lambert et al.

In this paper, we focus on the problem of integrating Energy-based Models (EBM) as guiding priors for motion optimization. EBMs are a set of neural networks that can represent expressive probability density distributions in terms of a Gibbs distribution parameterized by a suitable energy function. Due to their implicit nature, they can easily be integrated as optimization factors or as initial sampling distributions in the motion optimization problem, making them good candidates to integrate data-driven priors in the motion optimization problem. In this work, we present a set of required modeling and algorithmic choices to adapt EBMs into motion optimization. We investigate the benefit of including additional regularizers in the learning of the EBMs to use them with gradient-based optimizers and we present a set of EBM architectures to learn generalizable distributions for manipulation tasks. We present multiple cases in which the EBM could be integrated for motion optimization and evaluate the performance of learned EBMs as guiding priors for both simulated and real robot experiments.

Pascal Klink, Carlo D'Eramo, Jan Peters et al.

Curriculum reinforcement learning (CRL) allows solving complex tasks by generating a tailored sequence of learning tasks, starting from easy ones and subsequently increasing their difficulty. Although the potential of curricula in RL has been clearly shown in various works, it is less clear how to generate them for a given learning environment, resulting in various methods aiming to automate this task. In this work, we focus on framing curricula as interpolations between task distributions, which has previously been shown to be a viable approach to CRL. Identifying key issues of existing methods, we frame the generation of a curriculum as a constrained optimal transport problem between task distributions. Benchmarks show that this way of curriculum generation can improve upon existing CRL methods, yielding high performance in various tasks with different characteristics.

Marius Memmel, Puze Liu, Davide Tateo et al.

Black-box policy optimization is a class of reinforcement learning algorithms that explores and updates the policies at the parameter level. This class of algorithms is widely applied in robotics with movement primitives or non-differentiable policies. Furthermore, these approaches are particularly relevant where exploration at the action level could cause actuator damage or other safety issues. However, Black-box optimization does not scale well with the increasing dimensionality of the policy, leading to high demand for samples, which are expensive to obtain in real-world systems. In many practical applications, policy parameters do not contribute equally to the return. Identifying the most relevant parameters allows to narrow down the exploration and speed up the learning. Furthermore, updating only the effective parameters requires fewer samples, improving the scalability of the method. We present a novel method to prioritize the exploration of effective parameters and cope with full covariance matrix updates. Our algorithm learns faster than recent approaches and requires fewer samples to achieve state-of-the-art results. To select the effective parameters, we consider both the Pearson correlation coefficient and the Mutual Information. We showcase the capabilities of our approach on the Relative Entropy Policy Search algorithm in several simulated environments, including robotics simulations. Code is available at https://git.ias.informatik.tu-darmstadt.de/ias\_code/aistats2022/dr-creps}{git.ias.informatik.tu-darmstadt.de/ias\_code/aistats2022/dr-creps.

Tim Schneider, Boris Belousov, Hany Abdulsamad et al.

Robotic manipulation stands as a largely unsolved problem despite significant advances in robotics and machine learning in the last decades. One of the central challenges of manipulation is partial observability, as the agent usually does not know all physical properties of the environment and the objects it is manipulating in advance. A recently emerging theory that deals with partial observability in an explicit manner is Active Inference. It does so by driving the agent to act in a way that is not only goal-directed but also informative about the environment. In this work, we apply Active Inference to a hard-to-explore simulated robotic manipulation tasks, in which the agent has to balance a ball into a target zone. Since the reward of this task is sparse, in order to explore this environment, the agent has to learn to balance the ball without any extrinsic feedback, purely driven by its own curiosity. We show that the information-seeking behavior induced by Active Inference allows the agent to explore these challenging, sparse environments systematically. Finally, we conclude that using an information-seeking objective is beneficial in sparse environments and allows the agent to solve tasks in which methods that do not exhibit directed exploration fail.

Niklas Funk, Changqi Chen, Tim Schneider et al.

The field of robotic manipulation has advanced significantly in recent years. At the sensing level, several novel tactile sensors have been developed, capable of providing accurate contact information. On a methodological level, learning from demonstrations has proven an efficient paradigm to obtain performant robotic manipulation policies. The combination of both holds the promise to extract crucial contact-related information from the demonstration data and actively exploit it during policy rollouts. However, this integration has so far been underexplored, most notably in dynamic, contact-rich manipulation tasks where precision and reactivity are essential. This work therefore proposes a multimodal, visuotactile imitation learning framework that integrates a modular transformer architecture with a flow-based generative model, enabling efficient learning of fast and dexterous manipulation policies. We evaluate our framework on the dynamic, contact-rich task of robotic match lighting - a task in which tactile feedback influences human manipulation performance. The experimental results highlight the effectiveness of our approach and show that adding tactile information improves policy performance, thereby underlining their combined potential for learning dynamic manipulation from few demonstrations. Project website: https://sites.google.com/view/tactile-il .

Naqash Afzal, Niklas Funk, Erik Helmut et al.

Conventional robotic Braille readers typically rely on discrete, character-by-character scanning, limiting reading speed and disrupting natural flow. Vision-based alternatives often require substantial computation, introduce latency, and degrade in real-world conditions. In this work, we present a high accuracy, real-time pipeline for continuous Braille recognition using Evetac, an open-source neuromorphic event-based tactile sensor. Unlike frame-based vision systems, the neuromorphic tactile modality directly encodes dynamic contact events during continuous sliding, closely emulating human finger-scanning strategies. Our approach combines spatiotemporal segmentation with a lightweight ResNet-based classifier to process sparse event streams, enabling robust character recognition across varying indentation depths and scanning speeds. The proposed system achieves near-perfect accuracy (>=98%) at standard depths, generalizes across multiple Braille board layouts, and maintains strong performance under fast scanning. On a physical Braille board containing daily-living vocabulary, the system attains over 90% word-level accuracy, demonstrating robustness to temporal compression effects that challenge conventional methods. These results position neuromorphic tactile sensing as a scalable, low latency solution for robotic Braille reading, with broader implications for tactile perception in assistive and robotic applications.

Johanna Bethge, Maik Pfefferkorn, Alexander Rose et al.

We propose a model predictive control approach for autonomous vehicles that exploits learned Gaussian processes for predicting human driving behavior. The proposed approach employs the uncertainty about the GP's prediction to achieve safety. A multi-mode predictive control approach considers the possible intentions of the human drivers. While the intentions are represented by different Gaussian processes, their probabilities foreseen in the observed behaviors are determined by a suitable online classification. Intentions below a certain probability threshold are neglected to improve performance. The proposed multi-mode model predictive control approach with Gaussian process regression support enables repeated feasibility and probabilistic constraint satisfaction with high probability. The approach is underlined in simulation, considering real-world measurements for training the Gaussian processes.

Julen Urain, Ajay Mandlekar, Yilun Du et al.

Learning from Demonstrations, the field that proposes to learn robot behavior models from data, is gaining popularity with the emergence of deep generative models. Although the problem has been studied for years under names such as Imitation Learning, Behavioral Cloning, or Inverse Reinforcement Learning, classical methods have relied on models that don't capture complex data distributions well or don't scale well to large numbers of demonstrations. In recent years, the robot learning community has shown increasing interest in using deep generative models to capture the complexity of large datasets. In this survey, we aim to provide a unified and comprehensive review of the last year's progress in the use of deep generative models in robotics. We present the different types of models that the community has explored, such as energy-based models, diffusion models, action value maps, or generative adversarial networks. We also present the different types of applications in which deep generative models have been used, from grasp generation to trajectory generation or cost learning. One of the most important elements of generative models is the generalization out of distributions. In our survey, we review the different decisions the community has made to improve the generalization of the learned models. Finally, we highlight the research challenges and propose a number of future directions for learning deep generative models in robotics.

Stefan Löckel, Siwei Ju, Maximilian Schaller et al.

Engineering a high-performance race car requires a direct consideration of the human driver using real-world tests or Human-Driver-in-the-Loop simulations. Apart from that, offline simulations with human-like race driver models could make this vehicle development process more effective and efficient but are hard to obtain due to various challenges. With this work, we intend to provide a better understanding of race driver behavior and introduce an adaptive human race driver model based on imitation learning. Using existing findings and an interview with a professional race engineer, we identify fundamental adaptation mechanisms and how drivers learn to optimize lap time on a new track. Subsequently, we use these insights to develop generalization and adaptation techniques for a recently presented probabilistic driver modeling approach and evaluate it using data from professional race drivers and a state-of-the-art race car simulator. We show that our framework can create realistic driving line distributions on unseen race tracks with almost human-like performance. Moreover, our driver model optimizes its driving lap by lap, correcting driving errors from previous laps while achieving faster lap times. This work contributes to a better understanding and modeling of the human driver, aiming to expedite simulation methods in the modern vehicle development process and potentially supporting automated driving and racing technologies.

Andreas Look, Melih Kandemir, Barbara Rakitsch et al.

Many real-world dynamical systems can be described as State-Space Models (SSMs). In this formulation, each observation is emitted by a latent state, which follows first-order Markovian dynamics. A Probabilistic Deep SSM (ProDSSM) generalizes this framework to dynamical systems of unknown parametric form, where the transition and emission models are described by neural networks with uncertain weights. In this work, we propose the first deterministic inference algorithm for models of this type. Our framework allows efficient approximations for training and testing. We demonstrate in our experiments that our new method can be employed for a variety of tasks and enjoys a superior balance between predictive performance and computational budget.

Alexander I. Cowen-Rivers, Philip John Gorinski, Aivar Sootla et al.

Optimizing combinatorial structures is core to many real-world problems, such as those encountered in life sciences. For example, one of the crucial steps involved in antibody design is to find an arrangement of amino acids in a protein sequence that improves its binding with a pathogen. Combinatorial optimization of antibodies is difficult due to extremely large search spaces and non-linear objectives. Even for modest antibody design problems, where proteins have a sequence length of eleven, we are faced with searching over 2.05 x 10^14 structures. Applying traditional Reinforcement Learning algorithms such as Q-learning to combinatorial optimization results in poor performance. We propose Structured Q-learning (SQL), an extension of Q-learning that incorporates structural priors for combinatorial optimization. Using a molecular docking simulator, we demonstrate that SQL finds high binding energy sequences and performs favourably against baselines on eight challenging antibody design tasks, including designing antibodies for SARS-COV.

Joao Carvalho, An T. Le, Mark Baierl et al.

Learning priors on trajectory distributions can help accelerate robot motion planning optimization. Given previously successful plans, learning trajectory generative models as priors for a new planning problem is highly desirable. Prior works propose several ways on utilizing this prior to bootstrapping the motion planning problem. Either sampling the prior for initializations or using the prior distribution in a maximum-a-posterior formulation for trajectory optimization. In this work, we propose learning diffusion models as priors. We then can sample directly from the posterior trajectory distribution conditioned on task goals, by leveraging the inverse denoising process of diffusion models. Furthermore, diffusion has been recently shown to effectively encode data multimodality in high-dimensional settings, which is particularly well-suited for large trajectory dataset. To demonstrate our method efficacy, we compare our proposed method - Motion Planning Diffusion - against several baselines in simulated planar robot and 7-dof robot arm manipulator environments. To assess the generalization capabilities of our method, we test it in environments with previously unseen obstacles. Our experiments show that diffusion models are strong priors to encode high-dimensional trajectory distributions of robot motions.

Pascal Klink, Florian Wolf, Kai Ploeger et al.

Reinforcement Learning (RL) allows learning non-trivial robot control laws purely from data. However, many successful applications of RL have relied on ad-hoc regularizations, such as hand-crafted curricula, to regularize the learning performance. In this paper, we pair a recent algorithm for automatically building curricula with RL on massively parallelized simulations to learn a tracking controller for a spherical pendulum on a robotic arm via RL. Through an improved optimization scheme that better respects the non-Euclidean task structure, we allow the method to reliably generate curricula of trajectories to be tracked, resulting in faster and more robust learning compared to an RL baseline that does not exploit this form of structured learning. The learned policy matches the performance of an optimal control baseline on the real system, demonstrating the potential of curriculum RL to jointly learn state estimation and control for non-linear tracking tasks.

Michael Drolet, Simon Stepputtis, Siva Kailas et al.

Amidst the wide popularity of imitation learning algorithms in robotics, their properties regarding hyperparameter sensitivity, ease of training, data efficiency, and performance have not been well-studied in high-precision industry-inspired environments. In this work, we demonstrate the limitations and benefits of prominent imitation learning approaches and analyze their capabilities regarding these properties. We evaluate each algorithm on a complex bimanual manipulation task involving an over-constrained dynamics system in a setting involving multiple contacts between the manipulated object and the environment. While we find that imitation learning is well suited to solve such complex tasks, not all algorithms are equal in terms of handling environmental and hyperparameter perturbations, training requirements, performance, and ease of use. We investigate the empirical influence of these key characteristics by employing a carefully designed experimental procedure and learning environment. Paper website: https://bimanual-imitation.github.io/

Niklas Funk, Julen Urain, Joao Carvalho et al.

Spatial understanding is a critical aspect of most robotic tasks, particularly when generalization is important. Despite the impressive results of deep generative models in complex manipulation tasks, the absence of a representation that encodes intricate spatial relationships between observations and actions often limits spatial generalization, necessitating large amounts of demonstrations. To tackle this problem, we introduce a novel policy class, ActionFlow. ActionFlow integrates spatial symmetry inductive biases while generating expressive action sequences. On the representation level, ActionFlow introduces an SE(3) Invariant Transformer architecture, which enables informed spatial reasoning based on the relative SE(3) poses between observations and actions. For action generation, ActionFlow leverages Flow Matching, a state-of-the-art deep generative model known for generating high-quality samples with fast inference - an essential property for feedback control. In combination, ActionFlow policies exhibit strong spatial and locality biases and SE(3)-equivariant action generation. Our experiments demonstrate the effectiveness of ActionFlow and its two main components on several simulated and real-world robotic manipulation tasks and confirm that we can obtain equivariant, accurate, and efficient policies with spatially symmetric flow matching. Project website: https://flowbasedpolicies.github.io/

Carlos E. Luis, Alessandro G. Bottero, Julia Vinogradska et al.

We consider the problem of quantifying uncertainty over expected cumulative rewards in model-based reinforcement learning. In particular, we focus on characterizing the variance over values induced by a distribution over MDPs. Previous work upper bounds the posterior variance over values by solving a so-called uncertainty Bellman equation, but the over-approximation may result in inefficient exploration. We propose a new uncertainty Bellman equation whose solution converges to the true posterior variance over values and explicitly characterizes the gap in previous work. Moreover, our uncertainty quantification technique is easily integrated into common exploration strategies and scales naturally beyond the tabular setting by using standard deep reinforcement learning architectures. Experiments in difficult exploration tasks, both in tabular and continuous control settings, show that our sharper uncertainty estimates improve sample-efficiency.

Gabriele Tiboni, Pascal Klink, Jan Peters et al.

Varying dynamics parameters in simulation is a popular Domain Randomization (DR) approach for overcoming the reality gap in Reinforcement Learning (RL). Nevertheless, DR heavily hinges on the choice of the sampling distribution of the dynamics parameters, since high variability is crucial to regularize the agent's behavior but notoriously leads to overly conservative policies when randomizing excessively. In this paper, we propose a novel approach to address sim-to-real transfer, which automatically shapes dynamics distributions during training in simulation without requiring real-world data. We introduce DOmain RAndomization via Entropy MaximizatiON (DORAEMON), a constrained optimization problem that directly maximizes the entropy of the training distribution while retaining generalization capabilities. In achieving this, DORAEMON gradually increases the diversity of sampled dynamics parameters as long as the probability of success of the current policy is sufficiently high. We empirically validate the consistent benefits of DORAEMON in obtaining highly adaptive and generalizable policies, i.e. solving the task at hand across the widest range of dynamics parameters, as opposed to representative baselines from the DR literature. Notably, we also demonstrate the Sim2Real applicability of DORAEMON through its successful zero-shot transfer in a robotic manipulation setup under unknown real-world parameters.

Alessandro G. Bottero, Carlos E. Luis, Julia Vinogradska et al.

We consider a sequential decision making task where we are not allowed to evaluate parameters that violate an a priori unknown (safety) constraint. A common approach is to place a Gaussian process prior on the unknown constraint and allow evaluations only in regions that are safe with high probability. Most current methods rely on a discretization of the domain and cannot be directly extended to the continuous case. Moreover, the way in which they exploit regularity assumptions about the constraint introduces an additional critical hyperparameter. In this paper, we propose an information-theoretic safe exploration criterion that directly exploits the GP posterior to identify the most informative safe parameters to evaluate. Our approach is naturally applicable to continuous domains and does not require additional hyperparameters. We theoretically analyze the method and show that we do not violate the safety constraint with high probability and that we explore by learning about the constraint up to arbitrary precision. Empirical evaluations demonstrate improved data-efficiency and scalability.

Firas Al-Hafez, Guoping Zhao, Jan Peters et al.

Imitation Learning (IL) holds great promise for enabling agile locomotion in embodied agents. However, many existing locomotion benchmarks primarily focus on simplified toy tasks, often failing to capture the complexity of real-world scenarios and steering research toward unrealistic domains. To advance research in IL for locomotion, we present a novel benchmark designed to facilitate rigorous evaluation and comparison of IL algorithms. This benchmark encompasses a diverse set of environments, including quadrupeds, bipeds, and musculoskeletal human models, each accompanied by comprehensive datasets, such as real noisy motion capture data, ground truth expert data, and ground truth sub-optimal data, enabling evaluation across a spectrum of difficulty levels. To increase the robustness of learned agents, we provide an easy interface for dynamics randomization and offer a wide range of partially observable tasks to train agents across different embodiments. Finally, we provide handcrafted metrics for each task and ship our benchmark with state-of-the-art baseline algorithms to ease evaluation and enable fast benchmarking.

Firas Al-Hafez, Davide Tateo, Oleg Arenz et al.

Recent methods for imitation learning directly learn a $Q$-function using an implicit reward formulation rather than an explicit reward function. However, these methods generally require implicit reward regularization to improve stability and often mistreat absorbing states. Previous works show that a squared norm regularization on the implicit reward function is effective, but do not provide a theoretical analysis of the resulting properties of the algorithms. In this work, we show that using this regularizer under a mixture distribution of the policy and the expert provides a particularly illuminating perspective: the original objective can be understood as squared Bellman error minimization, and the corresponding optimization problem minimizes a bounded $χ^2$-Divergence between the expert and the mixture distribution. This perspective allows us to address instabilities and properly treat absorbing states. We show that our method, Least Squares Inverse Q-Learning (LS-IQ), outperforms state-of-the-art algorithms, particularly in environments with absorbing states. Finally, we propose to use an inverse dynamics model to learn from observations only. Using this approach, we retain performance in settings where no expert actions are available.

Ahmed Hendawy, Jan Peters, Carlo D'Eramo

Multi-Task Reinforcement Learning (MTRL) tackles the long-standing problem of endowing agents with skills that generalize across a variety of problems. To this end, sharing representations plays a fundamental role in capturing both unique and common characteristics of the tasks. Tasks may exhibit similarities in terms of skills, objects, or physical properties while leveraging their representations eases the achievement of a universal policy. Nevertheless, the pursuit of learning a shared set of diverse representations is still an open challenge. In this paper, we introduce a novel approach for representation learning in MTRL that encapsulates common structures among the tasks using orthogonal representations to promote diversity. Our method, named Mixture Of Orthogonal Experts (MOORE), leverages a Gram-Schmidt process to shape a shared subspace of representations generated by a mixture of experts. When task-specific information is provided, MOORE generates relevant representations from this shared subspace. We assess the effectiveness of our approach on two MTRL benchmarks, namely MiniGrid and MetaWorld, showing that MOORE surpasses related baselines and establishes a new state-of-the-art result on MetaWorld.

Daniel Palenicek, Michael Lutter, Joao Carvalho et al.

Model-based reinforcement learning is one approach to increase sample efficiency. However, the accuracy of the dynamics model and the resulting compounding error over modelled trajectories are commonly regarded as key limitations. A natural question to ask is: How much more sample efficiency can be gained by improving the learned dynamics models? Our paper empirically answers this question for the class of model-based value expansion methods in continuous control problems. Value expansion methods should benefit from increased model accuracy by enabling longer rollout horizons and better value function approximations. Our empirical study, which leverages oracle dynamics models to avoid compounding model errors, shows that (1) longer horizons increase sample efficiency, but the gain in improvement decreases with each additional expansion step, and (2) the increased model accuracy only marginally increases the sample efficiency compared to learned models with identical horizons. Therefore, longer horizons and increased model accuracy yield diminishing returns in terms of sample efficiency. These improvements in sample efficiency are particularly disappointing when compared to model-free value expansion methods. Even though they introduce no computational overhead, we find their performance to be on-par with model-based value expansion methods. Therefore, we conclude that the limitation of model-based value expansion methods is not the model accuracy of the learned models. While higher model accuracy is beneficial, our experiments show that even a perfect model will not provide an un-rivalled sample efficiency but that the bottleneck lies elsewhere.

Vignesh Prasad, Alap Kshirsagar, Dorothea Koert et al.

Shared dynamics models are important for capturing the complexity and variability inherent in Human-Robot Interaction (HRI). Therefore, learning such shared dynamics models can enhance coordination and adaptability to enable successful reactive interactions with a human partner. In this work, we propose a novel approach for learning a shared latent space representation for HRIs from demonstrations in a Mixture of Experts fashion for reactively generating robot actions from human observations. We train a Variational Autoencoder (VAE) to learn robot motions regularized using an informative latent space prior that captures the multimodality of the human observations via a Mixture Density Network (MDN). We show how our formulation derives from a Gaussian Mixture Regression formulation that is typically used approaches for learning HRI from demonstrations such as using an HMM/GMM for learning a joint distribution over the actions of the human and the robot. We further incorporate an additional regularization to prevent "mode collapse", a common phenomenon when using latent space mixture models with VAEs. We find that our approach of using an informative MDN prior from human observations for a VAE generates more accurate robot motions compared to previous HMM-based or recurrent approaches of learning shared latent representations, which we validate on various HRI datasets involving interactions such as handshakes, fistbumps, waving, and handovers. Further experiments in a real-world human-to-robot handover scenario show the efficacy of our approach for generating successful interactions with four different human interaction partners.

Aryaman Reddi, Maximilian Tölle, Jan Peters et al.

Robustness against adversarial attacks and distribution shifts is a long-standing goal of Reinforcement Learning (RL). To this end, Robust Adversarial Reinforcement Learning (RARL) trains a protagonist against destabilizing forces exercised by an adversary in a competitive zero-sum Markov game, whose optimal solution, i.e., rational strategy, corresponds to a Nash equilibrium. However, finding Nash equilibria requires facing complex saddle point optimization problems, which can be prohibitive to solve, especially for high-dimensional control. In this paper, we propose a novel approach for adversarial RL based on entropy regularization to ease the complexity of the saddle point optimization problem. We show that the solution of this entropy-regularized problem corresponds to a Quantal Response Equilibrium (QRE), a generalization of Nash equilibria that accounts for bounded rationality, i.e., agents sometimes play random actions instead of optimal ones. Crucially, the connection between the entropy-regularized objective and QRE enables free modulation of the rationality of the agents by simply tuning the temperature coefficient. We leverage this insight to propose our novel algorithm, Quantal Adversarial RL (QARL), which gradually increases the rationality of the adversary in a curriculum fashion until it is fully rational, easing the complexity of the optimization problem while retaining robustness. We provide extensive evidence of QARL outperforming RARL and recent baselines across several MuJoCo locomotion and navigation problems in overall performance and robustness.

Noah Farr, Aryaman Reddi, Carlo D'Eramo et al.

Streaming reinforcement learning has emerged as an online learning paradigm that conforms to the restrictions of natural learning agents that process data incrementally, i.e. with a batch size of 1 and no replay buffer. While streaming RL has recently been shown to scale with deep function approximation with full observability, partially observable settings have remained out of reach. Truncated backpropagation through time collapses to a one-step gradient horizon under the streaming setting, and exact real-time recurrent learning is prohibitively expensive. We close this gap using recurrent trace units, a diagonal recurrent architecture that enables exact RTRL with linear time and memory complexity in the parameter count, and show that they integrate cleanly into existing streaming algorithms across both discrete and continuous control. On a MemoryChain diagnostic with chain lengths from 2 to 128, our method sustains performance where streaming TBPTT(1) baselines using feedforward, GRU, and RTU networks collapse. On five POPGym tasks and on partially observable MuJoCo continuous control, the streaming approach is competitive with batched PPO on POPGym and recovers a substantial fraction of batched performance on masked MuJoCo, despite using no replay buffer or batched updates.

Hamish Flynn, David Reeb, Melih Kandemir et al.

PAC-Bayes has recently re-emerged as an effective theory with which one can derive principled learning algorithms with tight performance guarantees. However, applications of PAC-Bayes to bandit problems are relatively rare, which is a great misfortune. Many decision-making problems in healthcare, finance and natural sciences can be modelled as bandit problems. In many of these applications, principled algorithms with strong performance guarantees would be very much appreciated. This survey provides an overview of PAC-Bayes bounds for bandit problems and an experimental comparison of these bounds. On the one hand, we found that PAC-Bayes bounds are a useful tool for designing offline bandit algorithms with performance guarantees. In our experiments, a PAC-Bayesian offline contextual bandit algorithm was able to learn randomised neural network polices with competitive expected reward and non-vacuous performance guarantees. On the other hand, the PAC-Bayesian online bandit algorithms that we tested had loose cumulative regret bounds. We conclude by discussing some topics for future work on PAC-Bayesian bandit algorithms.

Hamish Flynn, David Reeb, Melih Kandemir et al.

We present a PAC-Bayesian analysis of lifelong learning. In the lifelong learning problem, a sequence of learning tasks is observed one-at-a-time, and the goal is to transfer information acquired from previous tasks to new learning tasks. We consider the case when each learning task is a multi-armed bandit problem. We derive lower bounds on the expected average reward that would be obtained if a given multi-armed bandit algorithm was run in a new task with a particular prior and for a set number of steps. We propose lifelong learning algorithms that use our new bounds as learning objectives. Our proposed algorithms are evaluated in several lifelong multi-armed bandit problems and are found to perform better than a baseline method that does not use generalisation bounds.

Nico Bohlinger, Grzegorz Czechmanowski, Maciej Krupka et al.

Deep Reinforcement Learning techniques are achieving state-of-the-art results in robust legged locomotion. While there exists a wide variety of legged platforms such as quadruped, humanoids, and hexapods, the field is still missing a single learning framework that can control all these different embodiments easily and effectively and possibly transfer, zero or few-shot, to unseen robot embodiments. We introduce URMA, the Unified Robot Morphology Architecture, to close this gap. Our framework brings the end-to-end Multi-Task Reinforcement Learning approach to the realm of legged robots, enabling the learned policy to control any type of robot morphology. The key idea of our method is to allow the network to learn an abstract locomotion controller that can be seamlessly shared between embodiments thanks to our morphology-agnostic encoders and decoders. This flexible architecture can be seen as a potential first step in building a foundation model for legged robot locomotion. Our experiments show that URMA can learn a locomotion policy on multiple embodiments that can be easily transferred to unseen robot platforms in simulation and the real world.

Daniel Palenicek, Michael Lutter, Jan Peters

Model-based value expansion methods promise to improve the quality of value function targets and, thereby, the effectiveness of value function learning. However, to date, these methods are being outperformed by Dyna-style algorithms with conceptually simpler 1-step value function targets. This shows that in practice, the theoretical justification of value expansion does not seem to hold. We provide a thorough empirical study to shed light on the causes of failure of value expansion methods in practice which is believed to be the compounding model error. By leveraging GPU based physics simulators, we are able to efficiently use the true dynamics for analysis inside the model-based reinforcement learning loop. Performing extensive comparisons between true and learned dynamics sheds light into this black box. This paper provides a better understanding of the actual problems in value expansion. We provide future directions of research by empirically testing the maximum theoretical performance of current approaches.