Adversarial Skill Learning for Robust Manipulation
This work addresses robustness issues in robotic manipulation for real-world applications, representing an incremental improvement over existing methods.
The paper tackles the problem of deep reinforcement learning policies failing under real-world disturbances in robotic manipulation by introducing adversarial training with a soft actor-critic algorithm, resulting in a policy robust to internal and external disturbances as validated in simulation and real-world experiments.
Deep reinforcement learning has made significant progress in robotic manipulation tasks and it works well in the ideal disturbance-free environment. However, in a real-world environment, both internal and external disturbances are inevitable, thus the performance of the trained policy will dramatically drop. To improve the robustness of the policy, we introduce the adversarial training mechanism to the robotic manipulation tasks in this paper, and an adversarial skill learning algorithm based on soft actor-critic (SAC) is proposed for robust manipulation. Extensive experiments are conducted to demonstrate that the learned policy is robust to internal and external disturbances. Additionally, the proposed algorithm is evaluated in both the simulation environment and on the real robotic platform.