Dieter Büchler

Simon Guist, Jan Schneider, Alexander Dittrich et al.

Reinforcement learning has shown great potential in solving complex tasks when large amounts of data can be generated with little effort. In robotics, one approach to generate training data builds on simulations based on dynamics models derived from first principles. However, for tasks that, for instance, involve complex soft robots, devising such models is substantially more challenging. Being able to train effectively in increasingly complicated scenarios with reinforcement learning enables to take advantage of complex systems such as soft robots. Here, we leverage the imbalance in complexity of the dynamics to learn more sample-efficiently. We (i) abstract the task into distinct components, (ii) off-load the simple dynamics parts into the simulation, and (iii) multiply these virtual parts to generate more data in hindsight. Our new method, Hindsight States (HiS), uses this data and selects the most useful transitions for training. It can be used with an arbitrary off-policy algorithm. We validate our method on several challenging simulated tasks and demonstrate that it improves learning both alone and when combined with an existing hindsight algorithm, Hindsight Experience Replay (HER). Finally, we evaluate HiS on a physical system and show that it boosts performance on a complex table tennis task with a muscular robot. Videos and code of the experiments can be found on webdav.tuebingen.mpg.de/his/.

Pierre Schumacher, Daniel Häufle, Dieter Büchler et al.

Muscle-actuated organisms are capable of learning an unparalleled diversity of dexterous movements despite their vast amount of muscles. Reinforcement learning (RL) on large musculoskeletal models, however, has not been able to show similar performance. We conjecture that ineffective exploration in large overactuated action spaces is a key problem. This is supported by the finding that common exploration noise strategies are inadequate in synthetic examples of overactuated systems. We identify differential extrinsic plasticity (DEP), a method from the domain of self-organization, as being able to induce state-space covering exploration within seconds of interaction. By integrating DEP into RL, we achieve fast learning of reaching and locomotion in musculoskeletal systems, outperforming current approaches in all considered tasks in sample efficiency and robustness.

Yi Zhao, Le Chen, Jan Schneider et al.

It has been a long-standing research goal to endow robot hands with human-level dexterity. Bi-manual robot piano playing constitutes a task that combines challenges from dynamic tasks, such as generating fast while precise motions, with slower but contact-rich manipulation problems. Although reinforcement learning based approaches have shown promising results in single-task performance, these methods struggle in a multi-song setting. Our work aims to close this gap and, thereby, enable imitation learning approaches for robot piano playing at scale. To this end, we introduce the Robot Piano 1 Million (RP1M) dataset, containing bi-manual robot piano playing motion data of more than one million trajectories. We formulate finger placements as an optimal transport problem, thus, enabling automatic annotation of vast amounts of unlabeled songs. Benchmarking existing imitation learning approaches shows that such approaches reach state-of-the-art robot piano playing performance by leveraging RP1M.

Isabell Wochner, Pierre Schumacher, Georg Martius et al.

Humans are able to outperform robots in terms of robustness, versatility, and learning of new tasks in a wide variety of movements. We hypothesize that highly nonlinear muscle dynamics play a large role in providing inherent stability, which is favorable to learning. While recent advances have been made in applying modern learning techniques to muscle-actuated systems both in simulation as well as in robotics, so far, no detailed analysis has been performed to show the benefits of muscles when learning from scratch. Our study closes this gap and showcases the potential of muscle actuators for core robotics challenges in terms of data-efficiency, hyperparameter sensitivity, and robustness.

Jan Schneider, Pierre Schumacher, Daniel Häufle et al.

Reinforcement learning~(RL) is a versatile framework for learning to solve complex real-world tasks. However, influences on the learning performance of RL algorithms are often poorly understood in practice. We discuss different analysis techniques and assess their effectiveness for investigating the impact of action representations in RL. Our experiments demonstrate that the action representation can significantly influence the learning performance on popular RL benchmark tasks. The analysis results indicate that some of the performance differences can be attributed to changes in the complexity of the optimization landscape. Finally, we discuss open challenges of analysis techniques for RL algorithms.

Jan Schneider, Pierre Schumacher, Simon Guist et al.

Policy gradient methods hold great potential for solving complex continuous control tasks. Still, their training efficiency can be improved by exploiting structure within the optimization problem. Recent work indicates that supervised learning can be accelerated by leveraging the fact that gradients lie in a low-dimensional and slowly-changing subspace. In this paper, we conduct a thorough evaluation of this phenomenon for two popular deep policy gradient methods on various simulated benchmark tasks. Our results demonstrate the existence of such gradient subspaces despite the continuously changing data distribution inherent to reinforcement learning. These findings reveal promising directions for future work on more efficient reinforcement learning, e.g., through improving parameter-space exploration or enabling second-order optimization.

Yi Zhao, Aidan Scannell, Wenshuai Zhao et al.

Leveraging offline data is a promising way to improve the sample efficiency of online reinforcement learning (RL). This paper expands the pool of usable data for offline-to-online RL by leveraging abundant non-curated data that is reward-free, of mixed quality, and collected across multiple embodiments. Although learning a world model appears promising for utilizing such data, we find that naive fine-tuning fails to accelerate RL training on many tasks. Through careful investigation, we attribute this failure to the distributional shift between offline and online data during fine-tuning. To address this issue and effectively use the offline data, we propose two essential techniques: \emph{i)} experience rehearsal and \emph{ii)} execution guidance. With these modifications, the non-curated offline data substantially improves RL's sample efficiency. Under limited sample budgets, our method achieves a 102.8\% relative improvement in aggregate score over learning-from-scratch baselines across 72 visuomotor tasks spanning 6 embodiments. On challenging tasks such as locomotion and robotic manipulation, it outperforms prior methods that utilize offline data by a decent margin.

Armin Kekić, Jan Schneider, Dieter Büchler et al.

Why do reinforcement learning (RL) policies fail or succeed? This is a challenging question due to the complex, high-dimensional nature of agent-environment interactions. In this work, we take a causal perspective on explaining the behavior of RL policies by viewing the states, actions, and rewards as variables in a low-level causal model. We introduce random perturbations to policy actions during execution and observe their effects on the cumulative reward, learning a simplified high-level causal model that explains these relationships. To this end, we develop a nonlinear Causal Model Reduction framework that ensures approximate interventional consistency, meaning the simplified high-level model responds to interventions in a similar way as the original complex system. We prove that for a class of nonlinear causal models, there exists a unique solution that achieves exact interventional consistency, ensuring learned explanations reflect meaningful causal patterns. Experiments on both synthetic causal models and practical RL tasks-including pendulum control and robot table tennis-demonstrate that our approach can uncover important behavioral patterns, biases, and failure modes in trained RL policies.

Nico Gürtler, Dieter Büchler, Georg Martius

Hierarchical reinforcement learning (HRL) holds great potential for sample-efficient learning on challenging long-horizon tasks. In particular, letting a higher level assign subgoals to a lower level has been shown to enable fast learning on difficult problems. However, such subgoal-based methods have been designed with static reinforcement learning environments in mind and consequently struggle with dynamic elements beyond the immediate control of the agent even though they are ubiquitous in real-world problems. In this paper, we introduce Hierarchical reinforcement learning with Timed Subgoals (HiTS), an HRL algorithm that enables the agent to adapt its timing to a dynamic environment by not only specifying what goal state is to be reached but also when. We discuss how communicating with a lower level in terms of such timed subgoals results in a more stable learning problem for the higher level. Our experiments on a range of standard benchmarks and three new challenging dynamic reinforcement learning environments show that our method is capable of sample-efficient learning where an existing state-of-the-art subgoal-based HRL method fails to learn stable solutions.

Dieter Büchler, Simon Guist, Roberto Calandra et al.

Dynamic tasks like table tennis are relatively easy to learn for humans but pose significant challenges to robots. Such tasks require accurate control of fast movements and precise timing in the presence of imprecise state estimation of the flying ball and the robot. Reinforcement Learning (RL) has shown promise in learning of complex control tasks from data. However, applying step-based RL to dynamic tasks on real systems is safety-critical as RL requires exploring and failing safely for millions of time steps in high-speed regimes. In this paper, we demonstrate that safe learning of table tennis using model-free Reinforcement Learning can be achieved by using robot arms driven by pneumatic artificial muscles (PAMs). Softness and back-drivability properties of PAMs prevent the system from leaving the safe region of its state space. In this manner, RL empowers the robot to return and smash real balls with 5 m\s and 12m\s on average to a desired landing point. Our setup allows the agent to learn this safety-critical task (i) without safety constraints in the algorithm, (ii) while maximizing the speed of returned balls directly in the reward function (iii) using a stochastic policy that acts directly on the low-level controls of the real system and (iv) trains for thousands of trials (v) from scratch without any prior knowledge. Additionally, we present HYSR, a practical hybrid sim and real training that avoids playing real balls during training by randomly replaying recorded ball trajectories in simulation and applying actions to the real robot. This work is the first to (a) fail-safe learn of a safety-critical dynamic task using anthropomorphic robot arms, (b) learn a precision-demanding problem with a PAM-driven system despite the control challenges and (c) train robots to play table tennis without real balls. Videos and datasets are available at muscularTT.embodied.ml.

Dieter Büchler, Roberto Calandra, Jan Peters

High-speed and high-acceleration movements are inherently hard to control. Applying learning to the control of such motions on anthropomorphic robot arms can improve the accuracy of the control but might damage the system. The inherent exploration of learning approaches can lead to instabilities and the robot reaching joint limits at high speeds. Having hardware that enables safe exploration of high-speed and high-acceleration movements is therefore desirable. To address this issue, we propose to use robots actuated by Pneumatic Artificial Muscles (PAMs). In this paper, we present a four degrees of freedom (DoFs) robot arm that reaches high joint angle accelerations of up to 28000 deg/s^2 while avoiding dangerous joint limits thanks to the antagonistic actuation and limits on the air pressure ranges. With this robot arm, we are able to tune control parameters using Bayesian optimization directly on the hardware without additional safety considerations. The achieved tracking performance on a fast trajectory exceeds previous results on comparable PAM-driven robots. We also show that our system can be controlled well on slow trajectories with PID controllers due to careful construction considerations such as minimal bending of cables, lightweight kinematics and minimal contact between PAMs and PAMs with the links. Finally, we propose a novel technique to control the the co-contraction of antagonistic muscle pairs. Experimental results illustrate that choosing the optimal co-contraction level is vital to reach better tracking performance. Through the use of PAM-driven robots and learning, we do a small step towards the future development of robots capable of more human-like motions.