This study proposes a new method, Wrench-Augmented Reinforcement Learning (WARL), which introduces a wrenche (force and torque) into the action space, and shows that introducing a wrench can encourage behaviors that do not sufficiently exploit the robot's physical embodiment.
Abstract
While reinforcement learning for legged robots has achieved high motor performance, it has been constrained by the limited exploration capability of actions confined to the joint space. To address this issue, this study proposes a new method, Wrench-Augmented Reinforcement Learning (WARL), which introduces a wrenche (force and torque) into the action space. The proposed method combines wrench-guided exploration with a success rate-based curriculum mechanism to expand exploration capabilities in the early stages of learning, with the ultimate goal of acquiring behaviors based solely on joint control. Experiments using a quadruped robot demonstrated that WARL can learn robustly across diverse terrains and motor tasks without requiring terrain-specific reward adjustments or complex curriculum designs. Furthermore, an ablation study verified the effectiveness of the Switching Curriculum, which gradually eliminates the wrench. On the other hand, we also show that introducing a wrench can encourage behaviors that do not sufficiently exploit the robot's physical embodiment. These findings suggest that while wrench-based exploration enhancement is effective for improving learning efficiency, designing it in a way that is consistent with the robot's physical structure is a critical future challenge.
This work proposes a hierarchical reinforcement learning pipeline that empowers the robots to perform aggressive locomotion through constrained obstacles--a narrow gate, extending the lifelike agility of legged robots to match that of their biological counterparts.
Zeren Luo, Jiahui Zhang, Yimin Han et al.· 1 citation
Effective learning design guidelines for realizing arm-based locomotion on life-sized robotic hardware and expanding the traversable workspace of robots are provided.
Ayumu Iwata, Kento Kawaharazuka, Keita Yoneda et al.· 0 citations
This study investigates the combination of a state-of-the-art reinforcement learning (RL) algorithm with human demonstrations to learn how to open a door with minimal task-specific engineering on an articulated soft robot arm and shows that combining LfD with RL results in both better performance and more robust behaviors.
Laurenz Elstner, Erik Kyrkjebø, M. Stoelen· Frontiers in Robotics and AI· 0 citations
Although reinforcement learning (RL) has shown great potential in legged robot motion control, traditional methods face two major bottlenecks: the strategy training of a single robot requires a large amount of computational resources, and the trained models are highly specific and complex for direct transfer and application. To address this key issue, this study proposes the Prior Transfer Reinforcement Learning (PTRL) framework, which improves training efficiency and model generalization ability through cross-robot knowledge transfer. The core contribution of this study is the construction of a three-stage learning paradigm of ‘pre-training, transfer, fine-tuning.’ Specifically, we first train the source robot strategy based on the Proximal Policy Optimization (PPO) algorithm and then achieve efficient knowledge transfer to the target robot by selectively freezing the key layers of the policy executor network. Through systematically designed multi-robot platform comparative experiments, it was verified that this method can reduce the training time, and the quantitative relationship between the proportion of frozen network layers and the transfer effect was revealed. The experimental results show that the proposed method significantly improves the performance of legged robots in walking tasks, demonstrating its strong applicability and advantages.
Haodong Huang, Shilong Sun, Hailin Huang et al.· Journal of Physics, Conferen...· 1 citation· ⚡1
This work identifies that additional research is still required to claim the successful resolution of the robotic arm reach-avoid task using DRL, and presents a comprehensive benchmark for the reachavoid task that accurately captures real-world complexities without simplifications.
Jonas Weihing, Shahram Eivazi· arXiv.org· 0 citations
Controlling high-dimensional musculoskeletal systems is challenging due to the large number of muscle actuators and the need for coordinated motor behavior. Recent work has shown that combining reinforcement learning with curriculum learning can improve performance on such tasks, yet the design of effective curricula remains an open question. In this work, we develop two curriculum strategies, Large-to-Small and Small-to-Large, against a no-curriculum baseline in order to control a musculoskeletal model of the arm during a task involving the tracing of geometric shapes. Both curriculum strategies significantly outperform the baseline, with success rates of 94% and 88%, in contrast to 42% without a curriculum. The best results are presented using the Large-to-Small strategy, which suggests that the agent learns better when it begins with larger geometrical shapes and gradually progresses on to smaller ones, where the tracing becomes more precise. Furthermore, we evaluate generalization to unseen instances, where the results demonstrate that curriculum-trained agents effectively generalize to novel shapes and scale variations, illustrating the robustness of the learned motor skills beyond the training conditions.The code for reproducing our experiments is accessible at https://github.com/AsmaeOUHSSAIN/CuRL-BioArm.git
Asmae Ouhssain, Mohamed-Harith Ibrahim, S. Harispe et al.· International Conference on...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
