Abhishek Gupta

H. J. Terry Suh, Glen Chou, Hongkai Dai et al. · mit

Gradient-based methods enable efficient search capabilities in high dimensions. However, in order to apply them effectively in offline optimization paradigms such as offline Reinforcement Learning (RL) or Imitation Learning (IL), we require a more careful consideration of how uncertainty estimation interplays with first-order methods that attempt to minimize them. We study smoothed distance to data as an uncertainty metric, and claim that it has two beneficial properties: (i) it allows gradient-based methods that attempt to minimize uncertainty to drive iterates to data as smoothing is annealed, and (ii) it facilitates analysis of model bias with Lipschitz constants. As distance to data can be expensive to compute online, we consider settings where we need amortize this computation. Instead of learning the distance however, we propose to learn its gradients directly as an oracle for first-order optimizers. We show these gradients can be efficiently learned with score-matching techniques by leveraging the equivalence between distance to data and data likelihood. Using this insight, we propose Score-Guided Planning (SGP), a planning algorithm for offline RL that utilizes score-matching to enable first-order planning in high-dimensional problems, where zeroth-order methods were unable to scale, and ensembles were unable to overcome local minima. Website: https://sites.google.com/view/score-guided-planning/home

Darshan Prabhu, Abhishek Gupta, Omkar Nitsure et al. · deepmind

Speech accents present a serious challenge to the performance of state-of-the-art end-to-end Automatic Speech Recognition (ASR) systems. Even with self-supervised learning and pre-training of ASR models, accent invariance is seldom achieved. In this work, we propose an accent-aware adaptation technique for self-supervised learning that introduces a trainable set of accent-specific codebooks to the self-supervised architecture. These learnable codebooks enable the model to capture accent specific information during pre-training, that is further refined during ASR finetuning. On the Mozilla Common Voice dataset, our proposed approach outperforms all other accent-adaptation approaches on both seen and unseen English accents, with up to 9% relative reduction in word error rate (WER).

Hao Chen, Abhishek Gupta, Yin Sun et al.

This paper studies Hoeffding's inequality for Markov chains under the generalized concentrability condition defined via integral probability metric (IPM). The generalized concentrability condition establishes a framework that interpolates and extends the existing hypotheses of Markov chain Hoeffding-type inequalities. The flexibility of our framework allows Hoeffding's inequality to be applied beyond the ergodic Markov chains in the traditional sense. We demonstrate the utility by applying our framework to several non-asymptotic analyses arising from the field of machine learning, including (i) a generalization bound for empirical risk minimization with Markovian samples, (ii) a finite sample guarantee for Ployak-Ruppert averaging of SGD, and (iii) a new regret bound for rested Markovian bandits with general state space.

Zhaoyi Zhou, Chuning Zhu, Runlong Zhou et al. · tsinghua

Off-policy dynamic programming (DP) techniques such as $Q$-learning have proven to be important in sequential decision-making problems. In the presence of function approximation, however, these techniques often diverge due to the absence of Bellman completeness in the function classes considered, a crucial condition for the success of DP-based methods. In this paper, we show how off-policy learning techniques based on return-conditioned supervised learning (RCSL) are able to circumvent these challenges of Bellman completeness, converging under significantly more relaxed assumptions inherited from supervised learning. We prove there exists a natural environment in which if one uses two-layer multilayer perceptron as the function approximator, the layer width needs to grow linearly with the state space size to satisfy Bellman completeness while a constant layer width is enough for RCSL. These findings take a step towards explaining the superior empirical performance of RCSL methods compared to DP-based methods in environments with near-optimal datasets. Furthermore, in order to learn from sub-optimal datasets, we propose a simple framework called MBRCSL, granting RCSL methods the ability of dynamic programming to stitch together segments from distinct trajectories. MBRCSL leverages learned dynamics models and forward sampling to accomplish trajectory stitching while avoiding the need for Bellman completeness that plagues all dynamic programming algorithms. We propose both theoretical analysis and experimental evaluation to back these claims, outperforming state-of-the-art model-free and model-based offline RL algorithms across several simulated robotics problems.

Zheyuan Hu, Aaron Rovinsky, Jianlan Luo et al.

Dexterous manipulation tasks involving contact-rich interactions pose a significant challenge for both model-based control systems and imitation learning algorithms. The complexity arises from the need for multi-fingered robotic hands to dynamically establish and break contacts, balance non-prehensile forces, and control large degrees of freedom. Reinforcement learning (RL) offers a promising approach due to its general applicability and capacity to autonomously acquire optimal manipulation strategies. However, its real-world application is often hindered by the necessity to generate a large number of samples, reset the environment, and obtain reward signals. In this work, we introduce an efficient system for learning dexterous manipulation skills with RL to alleviate these challenges. The main idea of our approach is the integration of recent advances in sample-efficient RL and replay buffer bootstrapping. This combination allows us to utilize data from different tasks or objects as a starting point for training new tasks, significantly improving learning efficiency. Additionally, our system completes the real-world training cycle by incorporating learned resets via an imitation-based pickup policy as well as learned reward functions, eliminating the need for manual resets and reward engineering. We demonstrate the benefits of reusing past data as replay buffer initialization for new tasks, for instance, the fast acquisition of intricate manipulation skills in the real world on a four-fingered robotic hand. (Videos: https://sites.google.com/view/reboot-dexterous)

Zoey Chen, Zhao Mandi, Homanga Bharadhwaj et al.

Generalization to unseen real-world scenarios for robot manipulation requires exposure to diverse datasets during training. However, collecting large real-world datasets is intractable due to high operational costs. For robot learning to generalize despite these challenges, it is essential to leverage sources of data or priors beyond the robot's direct experience. In this work, we posit that image-text generative models, which are pre-trained on large corpora of web-scraped data, can serve as such a data source. These generative models encompass a broad range of real-world scenarios beyond a robot's direct experience and can synthesize novel synthetic experiences that expose robotic agents to additional world priors aiding real-world generalization at no extra cost. In particular, our approach leverages pre-trained generative models as an effective tool for data augmentation. We propose a generative augmentation framework for semantically controllable augmentations and rapidly multiplying robot datasets while inducing rich variations that enable real-world generalization. Based on diverse augmentations of robot data, we show how scalable robot manipulation policies can be trained and deployed both in simulation and in unseen real-world environments such as kitchens and table-tops. By demonstrating the effectiveness of image-text generative models in diverse real-world robotic applications, our generative augmentation framework provides a scalable and efficient path for boosting generalization in robot learning at no extra human cost.

Chuning Zhu, Xinqi Wang, Tyler Han et al.

Intelligent agents must be generalists, capable of quickly adapting to various tasks. In reinforcement learning (RL), model-based RL learns a dynamics model of the world, in principle enabling transfer to arbitrary reward functions through planning. However, autoregressive model rollouts suffer from compounding error, making model-based RL ineffective for long-horizon problems. Successor features offer an alternative by modeling a policy's long-term state occupancy, reducing policy evaluation under new rewards to linear regression. Yet, zero-shot policy optimization for new tasks with successor features can be challenging. This work proposes a novel class of models, i.e., Distributional Successor Features for Zero-Shot Policy Optimization (DiSPOs), that learn a distribution of successor features of a stationary dataset's behavior policy, along with a policy that acts to realize different successor features achievable within the dataset. By directly modeling long-term outcomes in the dataset, DiSPOs avoid compounding error while enabling a simple scheme for zero-shot policy optimization across reward functions. We present a practical instantiation of DiSPOs using diffusion models and show their efficacy as a new class of transferable models, both theoretically and empirically across various simulated robotics problems. Videos and code available at https://weirdlabuw.github.io/dispo/.

Jayanth Reddy Regatti, Aniket Anand Deshmukh, Frank Cheng et al.

Offline reinforcement learning is used to train policies in scenarios where real-time access to the environment is expensive or impossible. As a natural consequence of these harsh conditions, an agent may lack the resources to fully observe the online environment before taking an action. We dub this situation the resource-constrained setting. This leads to situations where the offline dataset (available for training) can contain fully processed features (using powerful language models, image models, complex sensors, etc.) which are not available when actions are actually taken online. This disconnect leads to an interesting and unexplored problem in offline RL: Is it possible to use a richly processed offline dataset to train a policy which has access to fewer features in the online environment? In this work, we introduce and formalize this novel resource-constrained problem setting. We highlight the performance gap between policies trained using the full offline dataset and policies trained using limited features. We address this performance gap with a policy transfer algorithm which first trains a teacher agent using the offline dataset where features are fully available, and then transfers this knowledge to a student agent that only uses the resource-constrained features. To better capture the challenge of this setting, we propose a data collection procedure: Resource Constrained-Datasets for RL (RC-D4RL). We evaluate our transfer algorithm on RC-D4RL and the popular D4RL benchmarks and observe consistent improvement over the baseline (TD3+BC without transfer). The code for the experiments is available at https://github.com/JayanthRR/RC-OfflineRL.

Zoey Chen, Aaron Walsman, Marius Memmel et al.

Constructing simulation scenes that are both visually and physically realistic is a problem of practical interest in domains ranging from robotics to computer vision. This problem has become even more relevant as researchers wielding large data-hungry learning methods seek new sources of training data for physical decision-making systems. However, building simulation models is often still done by hand. A graphic designer and a simulation engineer work with predefined assets to construct rich scenes with realistic dynamic and kinematic properties. While this may scale to small numbers of scenes, to achieve the generalization properties that are required for data-driven robotic control, we require a pipeline that is able to synthesize large numbers of realistic scenes, complete with 'natural' kinematic and dynamic structures. To attack this problem, we develop models for inferring structure and generating simulation scenes from natural images, allowing for scalable scene generation from web-scale datasets. To train these image-to-simulation models, we show how controllable text-to-image generative models can be used in generating paired training data that allows for modeling of the inverse problem, mapping from realistic images back to complete scene models. We show how this paradigm allows us to build large datasets of scenes in simulation with semantic and physical realism. We present an integrated end-to-end pipeline that generates simulation scenes complete with articulated kinematic and dynamic structures from real-world images and use these for training robotic control policies. We then robustly deploy in the real world for tasks like articulated object manipulation. In doing so, our work provides both a pipeline for large-scale generation of simulation environments and an integrated system for training robust robotic control policies in the resulting environments.

Yi Li, Yuquan Deng, Jesse Zhang et al. · mit, nvidia

Large foundation models have shown strong open-world generalization to complex problems in vision and language, but similar levels of generalization have yet to be achieved in robotics. One fundamental challenge is the lack of robotic data, which are typically obtained through expensive on-robot operation. A promising remedy is to leverage cheaper, off-domain data such as action-free videos, hand-drawn sketches or simulation data. In this work, we posit that hierarchical vision-language-action (VLA) models can be more effective in utilizing off-domain data than standard monolithic VLA models that directly finetune vision-language models (VLMs) to predict actions. In particular, we study a class of hierarchical VLA models, where the high-level VLM is finetuned to produce a coarse 2D path indicating the desired robot end-effector trajectory given an RGB image and a task description. The intermediate 2D path prediction is then served as guidance to the low-level, 3D-aware control policy capable of precise manipulation. Doing so alleviates the high-level VLM from fine-grained action prediction, while reducing the low-level policy's burden on complex task-level reasoning. We show that, with the hierarchical design, the high-level VLM can transfer across significant domain gaps between the off-domain finetuning data and real-robot testing scenarios, including differences on embodiments, dynamics, visual appearances and task semantics, etc. In the real-robot experiments, we observe an average of 20% improvement in success rate across seven different axes of generalization over OpenVLA, representing a 50% relative gain. Visual results, code, and dataset are provided at: https://hamster-robot.github.io/

Marius Memmel, Andrew Wagenmaker, Chuning Zhu et al.

Model-free control strategies such as reinforcement learning have shown the ability to learn control strategies without requiring an accurate model or simulator of the world. While this is appealing due to the lack of modeling requirements, such methods can be sample inefficient, making them impractical in many real-world domains. On the other hand, model-based control techniques leveraging accurate simulators can circumvent these challenges and use a large amount of cheap simulation data to learn controllers that can effectively transfer to the real world. The challenge with such model-based techniques is the requirement for an extremely accurate simulation, requiring both the specification of appropriate simulation assets and physical parameters. This requires considerable human effort to design for every environment being considered. In this work, we propose a learning system that can leverage a small amount of real-world data to autonomously refine a simulation model and then plan an accurate control strategy that can be deployed in the real world. Our approach critically relies on utilizing an initial (possibly inaccurate) simulator to design effective exploration policies that, when deployed in the real world, collect high-quality data. We demonstrate the efficacy of this paradigm in identifying articulation, mass, and other physical parameters in several challenging robotic manipulation tasks, and illustrate that only a small amount of real-world data can allow for effective sim-to-real transfer. Project website at https://weirdlabuw.github.io/asid

Andrew Wagenmaker, Mitsuhiko Nakamoto, Yunchu Zhang et al.

Robotic control policies learned from human demonstrations have achieved impressive results in many real-world applications. However, in scenarios where initial performance is not satisfactory, as is often the case in novel open-world settings, such behavioral cloning (BC)-learned policies typically require collecting additional human demonstrations to further improve their behavior -- an expensive and time-consuming process. In contrast, reinforcement learning (RL) holds the promise of enabling autonomous online policy improvement, but often falls short of achieving this due to the large number of samples it typically requires. In this work we take steps towards enabling fast autonomous adaptation of BC-trained policies via efficient real-world RL. Focusing in particular on diffusion policies -- a state-of-the-art BC methodology -- we propose diffusion steering via reinforcement learning (DSRL): adapting the BC policy by running RL over its latent-noise space. We show that DSRL is highly sample efficient, requires only black-box access to the BC policy, and enables effective real-world autonomous policy improvement. Furthermore, DSRL avoids many of the challenges associated with finetuning diffusion policies, obviating the need to modify the weights of the base policy at all. We demonstrate DSRL on simulated benchmarks, real-world robotic tasks, and for adapting pretrained generalist policies, illustrating its sample efficiency and effective performance at real-world policy improvement.

Marius Memmel, Jacob Berg, Bingqing Chen et al.

Robot learning is witnessing a significant increase in the size, diversity, and complexity of pre-collected datasets, mirroring trends in domains such as natural language processing and computer vision. Many robot learning methods treat such datasets as multi-task expert data and learn a multi-task, generalist policy by training broadly across them. Notably, while these generalist policies can improve the average performance across many tasks, the performance of generalist policies on any one task is often suboptimal due to negative transfer between partitions of the data, compared to task-specific specialist policies. In this work, we argue for the paradigm of training policies during deployment given the scenarios they encounter: rather than deploying pre-trained policies to unseen problems in a zero-shot manner, we non-parametrically retrieve and train models directly on relevant data at test time. Furthermore, we show that many robotics tasks share considerable amounts of low-level behaviors and that retrieval at the "sub"-trajectory granularity enables significantly improved data utilization, generalization, and robustness in adapting policies to novel problems. In contrast, existing full-trajectory retrieval methods tend to underutilize the data and miss out on shared cross-task content. This work proposes STRAP, a technique for leveraging pre-trained vision foundation models and dynamic time warping to retrieve sub-sequences of trajectories from large training corpora in a robust fashion. STRAP outperforms both prior retrieval algorithms and multi-task learning methods in simulated and real experiments, showing the ability to scale to much larger offline datasets in the real world as well as the ability to learn robust control policies with just a handful of real-world demonstrations.

Hongchi Xia, Entong Su, Marius Memmel et al.

Creating virtual digital replicas from real-world data unlocks significant potential across domains like gaming and robotics. In this paper, we present DRAWER, a novel framework that converts a video of a static indoor scene into a photorealistic and interactive digital environment. Our approach centers on two main contributions: (i) a reconstruction module based on a dual scene representation that reconstructs the scene with fine-grained geometric details, and (ii) an articulation module that identifies articulation types and hinge positions, reconstructs simulatable shapes and appearances and integrates them into the scene. The resulting virtual environment is photorealistic, interactive, and runs in real time, with compatibility for game engines and robotic simulation platforms. We demonstrate the potential of DRAWER by using it to automatically create an interactive game in Unreal Engine and to enable real-to-sim-to-real transfer for robotics applications.

Patrick Yin, Tyler Westenbroek, Simran Bagaria et al.

Robot learning requires a considerable amount of high-quality data to realize the promise of generalization. However, large data sets are costly to collect in the real world. Physics simulators can cheaply generate vast data sets with broad coverage over states, actions, and environments. However, physics engines are fundamentally misspecified approximations to reality. This makes direct zero-shot transfer from simulation to reality challenging, especially in tasks where precise and force-sensitive manipulation is necessary. Thus, fine-tuning these policies with small real-world data sets is an appealing pathway for scaling robot learning. However, current reinforcement learning fine-tuning frameworks leverage general, unstructured exploration strategies which are too inefficient to make real-world adaptation practical. This paper introduces the Simulation-Guided Fine-tuning (SGFT) framework, which demonstrates how to extract structural priors from physics simulators to substantially accelerate real-world adaptation. Specifically, our approach uses a value function learned in simulation to guide real-world exploration. We demonstrate this approach across five real-world dexterous manipulation tasks where zero-shot sim-to-real transfer fails. We further demonstrate our framework substantially outperforms baseline fine-tuning methods, requiring up to an order of magnitude fewer real-world samples and succeeding at difficult tasks where prior approaches fail entirely. Last but not least, we provide theoretical justification for this new paradigm which underpins how SGFT can rapidly learn high-performance policies in the face of large sim-to-real dynamics gaps. Project webpage: https://weirdlabuw.github.io/sgft/{weirdlabuw.github.io/sgft}

Andrew Wagenmaker, Kevin Huang, Liyiming Ke et al.

In order to mitigate the sample complexity of real-world reinforcement learning, common practice is to first train a policy in a simulator where samples are cheap, and then deploy this policy in the real world, with the hope that it generalizes effectively. Such \emph{direct sim2real} transfer is not guaranteed to succeed, however, and in cases where it fails, it is unclear how to best utilize the simulator. In this work, we show that in many regimes, while direct sim2real transfer may fail, we can utilize the simulator to learn a set of \emph{exploratory} policies which enable efficient exploration in the real world. In particular, in the setting of low-rank MDPs, we show that coupling these exploratory policies with simple, practical approaches -- least-squares regression oracles and naive randomized exploration -- yields a polynomial sample complexity in the real world, an exponential improvement over direct sim2real transfer, or learning without access to a simulator. To the best of our knowledge, this is the first evidence that simulation transfer yields a provable gain in reinforcement learning in settings where direct sim2real transfer fails. We validate our theoretical results on several realistic robotic simulators and a real-world robotic sim2real task, demonstrating that transferring exploratory policies can yield substantial gains in practice as well.

Mateo Guaman Castro, Sidharth Rajagopal, Daniel Gorbatov et al.

A fundamental challenge in robot navigation lies in learning policies that generalize across diverse environments while conforming to the unique physical constraints and capabilities of a specific embodiment (e.g., quadrupeds can walk up stairs, but rovers cannot). We propose VAMOS, a hierarchical VLA that decouples semantic planning from embodiment grounding: a generalist planner learns from diverse, open-world data, while a specialist affordance model learns the robot's physical constraints and capabilities in safe, low-cost simulation. We enabled this separation by carefully designing an interface that lets a high-level planner propose candidate paths directly in image space that the affordance model then evaluates and re-ranks. Our real-world experiments show that VAMOS achieves higher success rates in both indoor and complex outdoor navigation than state-of-the-art model-based and end-to-end learning methods. We also show that our hierarchical design enables cross-embodied navigation across legged and wheeled robots and is easily steerable using natural language. Real-world ablations confirm that the specialist model is key to embodiment grounding, enabling a single high-level planner to be deployed across physically distinct wheeled and legged robots. Finally, this model significantly enhances single-robot reliability, achieving 3X higher success rates by rejecting physically infeasible plans. Website: https://vamos-vla.github.io/

Elie Aljalbout, Jiaxu Xing, Angel Romero et al. · mit, nvidia

Machine learning has facilitated significant advancements across various robotics domains, including navigation, locomotion, and manipulation. Many such achievements have been driven by the extensive use of simulation as a critical tool for training and testing robotic systems prior to their deployment in real-world environments. However, simulations consist of abstractions and approximations that inevitably introduce discrepancies between simulated and real environments, known as the reality gap. These discrepancies significantly hinder the successful transfer of systems from simulation to the real world. Closing this gap remains one of the most pressing challenges in robotics. Recent advances in sim-to-real transfer have demonstrated promising results across various platforms, including locomotion, navigation, and manipulation. By leveraging techniques such as domain randomization, real-to-sim transfer, state and action abstractions, and sim-real co-training, many works have overcome the reality gap. However, challenges persist, and a deeper understanding of the reality gap's root causes and solutions is necessary. In this survey, we present a comprehensive overview of the sim-to-real landscape, highlighting the causes, solutions, and evaluation metrics for the reality gap and sim-to-real transfer.

Abhishek Gupta, Amruta Parulekar, Sameep Chattopadhyay et al.

Spontaneous or conversational multilingual speech presents many challenges for state-of-the-art automatic speech recognition (ASR) systems. In this work, we present a new technique AMPS that augments a multilingual multimodal ASR system with paraphrase-based supervision for improved conversational ASR in multiple languages, including Hindi, Marathi, Malayalam, Kannada, and Nyanja. We use paraphrases of the reference transcriptions as additional supervision while training the multimodal ASR model and selectively invoke this paraphrase objective for utterances with poor ASR performance. Using AMPS with a state-of-the-art multimodal model SeamlessM4T, we obtain significant relative reductions in word error rates (WERs) of up to 5%. We present detailed analyses of our system using both objective and human evaluation metrics.

Abhishek Gupta, Amruta Parulekar, Sameep Chattopadhyay et al.

Automatic speech recognition (ASR) for low-resource languages remains a challenge due to the scarcity of labeled training data. Parameter-efficient fine-tuning and text-only adaptation are two popular methods that have been used to address such low-resource settings. In this work, we investigate how these techniques can be effectively combined using a multilingual multimodal model like SeamlessM4T. Multimodal models are able to leverage unlabeled text via text-only adaptation with further parameter-efficient ASR fine-tuning, thus boosting ASR performance. We also show cross-lingual transfer from a high-resource language, achieving up to a relative 17% WER reduction over a baseline in a zero-shot setting without any labeled speech.

Hao Chen, Jiacheng Tang, Abhishek Gupta

In this paper, we develop a new change detection algorithm for detecting a change in the Markov kernel over a metric space in which the post-change kernel is unknown. Under the assumption that the pre- and post-change Markov kernel is uniformly ergodic, we derive an upper bound on the mean delay and a lower bound on the mean time between false alarms. A numerical simulation is provided to demonstrate the effectiveness of our method.

Jian Cheng Wong, Chinchun Ooi, Abhishek Gupta et al.

A physics-informed neural network (PINN) uses physics-augmented loss functions, e.g., incorporating the residual term from governing partial differential equations (PDEs), to ensure its output is consistent with fundamental physics laws. However, it turns out to be difficult to train an accurate PINN model for many problems in practice. In this paper, we present a novel perspective of the merits of learning in sinusoidal spaces with PINNs. By analyzing behavior at model initialization, we first show that a PINN of increasing expressiveness induces an initial bias around flat output functions. Notably, this initial solution can be very close to satisfying many physics PDEs, i.e., falling into a local minimum of the PINN loss that only minimizes PDE residuals, while still being far from the true solution that jointly minimizes PDE residuals and the initial and/or boundary conditions. It is difficult for gradient descent optimization to escape from such a local minimum trap, often causing the training to stall. We then prove that the sinusoidal mapping of inputs, in an architecture we label as sf-PINN, is effective to increase input gradient variability, thus avoiding being trapped in such deceptive local minimum. The level of variability can be effectively modulated to match high-frequency patterns in the problem at hand. A key facet of this paper is the comprehensive empirical study that demonstrates the efficacy of learning in sinusoidal spaces with PINNs for a wide range of forward and inverse modelling problems spanning multiple physics domains.

Archit Sharma, Abhishek Gupta, Sergey Levine et al.

Reinforcement learning (RL) promises to enable autonomous acquisition of complex behaviors for diverse agents. However, the success of current reinforcement learning algorithms is predicated on an often under-emphasised requirement -- each trial needs to start from a fixed initial state distribution. Unfortunately, resetting the environment to its initial state after each trial requires substantial amount of human supervision and extensive instrumentation of the environment which defeats the goal of autonomous acquisition of complex behaviors. In this work, we propose Value-accelerated Persistent Reinforcement Learning (VaPRL), which generates a curriculum of initial states such that the agent can bootstrap on the success of easier tasks to efficiently learn harder tasks. The agent also learns to reach the initial states proposed by the curriculum, minimizing the reliance on human interventions into the learning. We observe that VaPRL reduces the interventions required by three orders of magnitude compared to episodic RL while outperforming prior state-of-the art methods for reset-free RL both in terms of sample efficiency and asymptotic performance on a variety of simulated robotics problems.

Kevin Li, Abhishek Gupta, Ashwin Reddy et al.

Exploration in reinforcement learning is a challenging problem: in the worst case, the agent must search for high-reward states that could be hidden anywhere in the state space. Can we define a more tractable class of RL problems, where the agent is provided with examples of successful outcomes? In this problem setting, the reward function can be obtained automatically by training a classifier to categorize states as successful or not. If trained properly, such a classifier can provide a well-shaped objective landscape that both promotes progress toward good states and provides a calibrated exploration bonus. In this work, we show that an uncertainty aware classifier can solve challenging reinforcement learning problems by both encouraging exploration and provided directed guidance towards positive outcomes. We propose a novel mechanism for obtaining these calibrated, uncertainty-aware classifiers based on an amortized technique for computing the normalized maximum likelihood (NML) distribution. To make this tractable, we propose a novel method for computing the NML distribution by using meta-learning. We show that the resulting algorithm has a number of intriguing connections to both count-based exploration methods and prior algorithms for learning reward functions, while also providing more effective guidance towards the goal. We demonstrate that our algorithm solves a number of challenging navigation and robotic manipulation tasks which prove difficult or impossible for prior methods.

Jayanth Regatti, Hao Chen, Abhishek Gupta

We propose two novel stochastic gradient descent algorithms, ByGARS and ByGARS++, for distributed machine learning in the presence of any number of Byzantine adversaries. In these algorithms, reputation scores of workers are computed using an auxiliary dataset at the server. This reputation score is then used for aggregating the gradients for stochastic gradient descent. The computational complexity of ByGARS++ is the same as the usual distributed stochastic gradient descent method with only an additional inner product computation in every iteration. We show that using these reputation scores for gradient aggregation is robust to any number of multiplicative noise Byzantine adversaries and use two-timescale stochastic approximation theory to prove convergence for strongly convex loss functions. We demonstrate the effectiveness of the algorithms for non-convex learning problems using MNIST and CIFAR-10 datasets against almost all state-of-the-art Byzantine attacks. We also show that the proposed algorithms are robust to multiple different types of attacks at the same time.

Abhishek Gupta, Camylle Lanteigne, Sara Kingsley

In a world increasingly dominated by AI applications, an understudied aspect is the carbon and social footprint of these power-hungry algorithms that require copious computation and a trove of data for training and prediction. While profitable in the short-term, these practices are unsustainable and socially extractive from both a data-use and energy-use perspective. This work proposes an ESG-inspired framework combining socio-technical measures to build eco-socially responsible AI systems. The framework has four pillars: compute-efficient machine learning, federated learning, data sovereignty, and a LEEDesque certificate. Compute-efficient machine learning is the use of compressed network architectures that show marginal decreases in accuracy. Federated learning augments the first pillar's impact through the use of techniques that distribute computational loads across idle capacity on devices. This is paired with the third pillar of data sovereignty to ensure the privacy of user data via techniques like use-based privacy and differential privacy. The final pillar ties all these factors together and certifies products and services in a standardized manner on their environmental and social impacts, allowing consumers to align their purchase with their values.

Henry Zhu, Justin Yu, Abhishek Gupta et al.

The success of reinforcement learning for real world robotics has been, in many cases limited to instrumented laboratory scenarios, often requiring arduous human effort and oversight to enable continuous learning. In this work, we discuss the elements that are needed for a robotic learning system that can continually and autonomously improve with data collected in the real world. We propose a particular instantiation of such a system, using dexterous manipulation as our case study. Subsequently, we investigate a number of challenges that come up when learning without instrumentation. In such settings, learning must be feasible without manually designed resets, using only on-board perception, and without hand-engineered reward functions. We propose simple and scalable solutions to these challenges, and then demonstrate the efficacy of our proposed system on a set of dexterous robotic manipulation tasks, providing an in-depth analysis of the challenges associated with this learning paradigm. We demonstrate that our complete system can learn without any human intervention, acquiring a variety of vision-based skills with a real-world three-fingered hand. Results and videos can be found at https://sites.google.com/view/realworld-rl/

Aviral Kumar, Abhishek Gupta, Sergey Levine

Deep reinforcement learning can learn effective policies for a wide range of tasks, but is notoriously difficult to use due to instability and sensitivity to hyperparameters. The reasons for this remain unclear. When using standard supervised methods (e.g., for bandits), on-policy data collection provides "hard negatives" that correct the model in precisely those states and actions that the policy is likely to visit. We call this phenomenon "corrective feedback." We show that bootstrapping-based Q-learning algorithms do not necessarily benefit from this corrective feedback, and training on the experience collected by the algorithm is not sufficient to correct errors in the Q-function. In fact, Q-learning and related methods can exhibit pathological interactions between the distribution of experience collected by the agent and the policy induced by training on that experience, leading to potential instability, sub-optimal convergence, and poor results when learning from noisy, sparse or delayed rewards. We demonstrate the existence of this problem, both theoretically and empirically. We then show that a specific correction to the data distribution can mitigate this issue. Based on these observations, we propose a new algorithm, DisCor, which computes an approximation to this optimal distribution and uses it to re-weight the transitions used for training, resulting in substantial improvements in a range of challenging RL settings, such as multi-task learning and learning from noisy reward signals. Blog post presenting a summary of this work is available at: https://bair.berkeley.edu/blog/2020/03/16/discor/.

Rahul Singh, Abhishek Gupta, Ness B. Shroff

We consider reinforcement learning (RL) in Markov Decision Processes in which an agent repeatedly interacts with an environment that is modeled by a controlled Markov process. At each time step $t$, it earns a reward, and also incurs a cost-vector consisting of $M$ costs. We design model-based RL algorithms that maximize the cumulative reward earned over a time horizon of $T$ time-steps, while simultaneously ensuring that the average values of the $M$ cost expenditures are bounded by agent-specified thresholds $c^{ub}_i,i=1,2,\ldots,M$. In order to measure the performance of a reinforcement learning algorithm that satisfies the average cost constraints, we define an $M+1$ dimensional regret vector that is composed of its reward regret, and $M$ cost regrets. The reward regret measures the sub-optimality in the cumulative reward, while the $i$-th component of the cost regret vector is the difference between its $i$-th cumulative cost expense and the expected cost expenditures $Tc^{ub}_i$. We prove that the expected value of the regret vector of UCRL-CMDP, is upper-bounded as $\tilde{O}\left(T^{2\slash 3}\right)$, where $T$ is the time horizon. We further show how to reduce the regret of a desired subset of the $M$ costs, at the expense of increasing the regrets of rewards and the remaining costs. To the best of our knowledge, ours is the only work that considers non-episodic RL under average cost constraints, and derive algorithms that can~\emph{tune the regret vector} according to the agent's requirements on its cost regrets.

Abhishek Gupta, Vikash Kumar, Corey Lynch et al.

We present relay policy learning, a method for imitation and reinforcement learning that can solve multi-stage, long-horizon robotic tasks. This general and universally-applicable, two-phase approach consists of an imitation learning stage that produces goal-conditioned hierarchical policies, and a reinforcement learning phase that finetunes these policies for task performance. Our method, while not necessarily perfect at imitation learning, is very amenable to further improvement via environment interaction, allowing it to scale to challenging long-horizon tasks. We simplify the long-horizon policy learning problem by using a novel data-relabeling algorithm for learning goal-conditioned hierarchical policies, where the low-level only acts for a fixed number of steps, regardless of the goal achieved. While we rely on demonstration data to bootstrap policy learning, we do not assume access to demonstrations of every specific tasks that is being solved, and instead leverage unstructured and unsegmented demonstrations of semantically meaningful behaviors that are not only less burdensome to provide, but also can greatly facilitate further improvement using reinforcement learning. We demonstrate the effectiveness of our method on a number of multi-stage, long-horizon manipulation tasks in a challenging kitchen simulation environment. Videos are available at https://relay-policy-learning.github.io/

Jayanth Regatti, Gaurav Tendolkar, Yi Zhou et al.

The performance of fully synchronized distributed systems has faced a bottleneck due to the big data trend, under which asynchronous distributed systems are becoming a major popularity due to their powerful scalability. In this paper, we study the generalization performance of stochastic gradient descent (SGD) on a distributed asynchronous system. The system consists of multiple worker machines that compute stochastic gradients which are further sent to and aggregated on a common parameter server to update the variables, and the communication in the system suffers from possible delays. Under the algorithm stability framework, we prove that distributed asynchronous SGD generalizes well given enough data samples in the training optimization. In particular, our results suggest to reduce the learning rate as we allow more asynchrony in the distributed system. Such adaptive learning rate strategy improves the stability of the distributed algorithm and reduces the corresponding generalization error. Then, we confirm our theoretical findings via numerical experiments.

Xinyi Ren, Jianlan Luo, Eugen Solowjow et al.

Accurate state estimation is a fundamental component of robotic control. In robotic manipulation tasks, as is our focus in this work, state estimation is essential for identifying the positions of objects in the scene, forming the basis of the manipulation plan. However, pose estimation typically requires expensive 3D cameras or additional instrumentation such as fiducial markers to perform accurately. Recently, Tobin et al.~introduced an approach to pose estimation based on domain randomization, where a neural network is trained to predict pose directly from a 2D image of the scene. The network is trained on computer-generated images with a high variation in textures and lighting, thereby generalizing to real-world images. In this work, we investigate how to improve the accuracy of domain randomization based pose estimation. Our main idea is that active perception -- moving the robot to get a better estimate of pose -- can be trained in simulation and transferred to real using domain randomization. In our approach, the robot trains in a domain-randomized simulation how to estimate pose from a \emph{sequence} of images. We show that our approach can significantly improve the accuracy of standard pose estimation in several scenarios: when the robot holding an object moves, when reference objects are moved in the scene, or when the camera is moved around the object.

Tuomas Haarnoja, Aurick Zhou, Kristian Hartikainen et al.

Model-free deep reinforcement learning (RL) algorithms have been successfully applied to a range of challenging sequential decision making and control tasks. However, these methods typically suffer from two major challenges: high sample complexity and brittleness to hyperparameters. Both of these challenges limit the applicability of such methods to real-world domains. In this paper, we describe Soft Actor-Critic (SAC), our recently introduced off-policy actor-critic algorithm based on the maximum entropy RL framework. In this framework, the actor aims to simultaneously maximize expected return and entropy. That is, to succeed at the task while acting as randomly as possible. We extend SAC to incorporate a number of modifications that accelerate training and improve stability with respect to the hyperparameters, including a constrained formulation that automatically tunes the temperature hyperparameter. We systematically evaluate SAC on a range of benchmark tasks, as well as real-world challenging tasks such as locomotion for a quadrupedal robot and robotic manipulation with a dexterous hand. With these improvements, SAC achieves state-of-the-art performance, outperforming prior on-policy and off-policy methods in sample-efficiency and asymptotic performance. Furthermore, we demonstrate that, in contrast to other off-policy algorithms, our approach is very stable, achieving similar performance across different random seeds. These results suggest that SAC is a promising candidate for learning in real-world robotics tasks.

Henry Zhu, Abhishek Gupta, Aravind Rajeswaran et al.

Dexterous multi-fingered robotic hands can perform a wide range of manipulation skills, making them an appealing component for general-purpose robotic manipulators. However, such hands pose a major challenge for autonomous control, due to the high dimensionality of their configuration space and complex intermittent contact interactions. In this work, we propose deep reinforcement learning (deep RL) as a scalable solution for learning complex, contact rich behaviors with multi-fingered hands. Deep RL provides an end-to-end approach to directly map sensor readings to actions, without the need for task specific models or policy classes. We show that contact-rich manipulation behavior with multi-fingered hands can be learned by directly training with model-free deep RL algorithms in the real world, with minimal additional assumption and without the aid of simulation. We learn a variety of complex behaviors on two different low-cost hardware platforms. We show that each task can be learned entirely from scratch, and further study how the learning process can be further accelerated by using a small number of human demonstrations to bootstrap learning. Our experiments demonstrate that complex multi-fingered manipulation skills can be learned in the real world in about 4-7 hours for most tasks, and that demonstrations can decrease this to 2-3 hours, indicating that direct deep RL training in the real world is a viable and practical alternative to simulation and model-based control. \url{https://sites.google.com/view/deeprl-handmanipulation}

YuXuan Liu, Abhishek Gupta, Pieter Abbeel et al.

Imitation learning is an effective approach for autonomous systems to acquire control policies when an explicit reward function is unavailable, using supervision provided as demonstrations from an expert, typically a human operator. However, standard imitation learning methods assume that the agent receives examples of observation-action tuples that could be provided, for instance, to a supervised learning algorithm. This stands in contrast to how humans and animals imitate: we observe another person performing some behavior and then figure out which actions will realize that behavior, compensating for changes in viewpoint, surroundings, object positions and types, and other factors. We term this kind of imitation learning "imitation-from-observation," and propose an imitation learning method based on video prediction with context translation and deep reinforcement learning. This lifts the assumption in imitation learning that the demonstration should consist of observations in the same environment configuration, and enables a variety of interesting applications, including learning robotic skills that involve tool use simply by observing videos of human tool use. Our experimental results show the effectiveness of our approach in learning a wide range of real-world robotic tasks modeled after common household chores from videos of a human demonstrator, including sweeping, ladling almonds, pushing objects as well as a number of tasks in simulation.

Abhishek Gupta, Coline Devin, YuXuan Liu et al.

People can learn a wide range of tasks from their own experience, but can also learn from observing other creatures. This can accelerate acquisition of new skills even when the observed agent differs substantially from the learning agent in terms of morphology. In this paper, we examine how reinforcement learning algorithms can transfer knowledge between morphologically different agents (e.g., different robots). We introduce a problem formulation where two agents are tasked with learning multiple skills by sharing information. Our method uses the skills that were learned by both agents to train invariant feature spaces that can then be used to transfer other skills from one agent to another. The process of learning these invariant feature spaces can be viewed as a kind of "analogy making", or implicit learning of partial correspondences between two distinct domains. We evaluate our transfer learning algorithm in two simulated robotic manipulation skills, and illustrate that we can transfer knowledge between simulated robotic arms with different numbers of links, as well as simulated arms with different actuation mechanisms, where one robot is torque-driven while the other is tendon-driven.

Vikash Kumar, Abhishek Gupta, Emanuel Todorov et al.

We explore learning-based approaches for feedback control of a dexterous five-finger hand performing non-prehensile manipulation. First, we learn local controllers that are able to perform the task starting at a predefined initial state. These controllers are constructed using trajectory optimization with respect to locally-linear time-varying models learned directly from sensor data. In some cases, we initialize the optimizer with human demonstrations collected via teleoperation in a virtual environment. We demonstrate that such controllers can perform the task robustly, both in simulation and on the physical platform, for a limited range of initial conditions around the trained starting state. We then consider two interpolation methods for generalizing to a wider range of initial conditions: deep learning, and nearest neighbors. We find that nearest neighbors achieve higher performance. Nevertheless, the neural network has its advantages: it uses only tactile and proprioceptive feedback but no visual feedback about the object (i.e. it performs the task blind) and learns a time-invariant policy. In contrast, the nearest neighbors method switches between time-varying local controllers based on the proximity of initial object states sensed via motion capture. While both generalization methods leave room for improvement, our work shows that (i) local trajectory-based controllers for complex non-prehensile manipulation tasks can be constructed from surprisingly small amounts of training data, and (ii) collections of such controllers can be interpolated to form more global controllers. Results are summarized in the supplementary video: https://youtu.be/E0wmO6deqjo

Abhishek Gupta, Clemens Eppner, Sergey Levine et al.

Dexterous multi-fingered hands can accomplish fine manipulation behaviors that are infeasible with simple robotic grippers. However, sophisticated multi-fingered hands are often expensive and fragile. Low-cost soft hands offer an appealing alternative to more conventional devices, but present considerable challenges in sensing and actuation, making them difficult to apply to more complex manipulation tasks. In this paper, we describe an approach to learning from demonstration that can be used to train soft robotic hands to perform dexterous manipulation tasks. Our method uses object-centric demonstrations, where a human demonstrates the desired motion of manipulated objects with their own hands, and the robot autonomously learns to imitate these demonstrations using reinforcement learning. We propose a novel algorithm that allows us to blend and select a subset of the most feasible demonstrations to learn to imitate on the hardware, which we use with an extension of the guided policy search framework to use multiple demonstrations to learn generalizable neural network policies. We demonstrate our approach on the RBO Hand 2, with learned motor skills for turning a valve, manipulating an abacus, and grasping.