Alexander Panov Konstantin Mironov Andrey Gorodetsky

Model-Based Policy Optimization with Neural Differential Equations for Robotic Arm Control


Applying learning-based control methods to real robots presents hard challenges, including the low sample efficiency of model-free reinforcement learning algorithms. The widely adopted approach to tackling this problem uses an environment dynamics model. We propose to use the Neural Ordinary Differential Equations to approximate transition dynamics as this allows for finer control of a trajectory generation process. NODE offers a continuous-time formulation that captures the temporal dependencies. We evaluate our approach on various tasks from simulation environment including learning 6-DoF robotic arm to open the door, which represents particular challenges for policy search. The NODE model is trained to predict movement of the arm and the door, and is used to generate trajectories for the model-based policy optimization. Our method shows better sample efficiency on this task comparing to the model-free and model-based baseline. It also shows comparable results on several other tasks. The application of NODE to model-based reinforcement learning enables more precise modeling of robotic system dynamics and enhances the sample efficiency of learning-based control methods. The empirical evaluation on various tasks demonstrates the efficacy of our approach, offering promising prospects for improving the performance and efficiency of real-world robotic systems.