Enforcing the consensus between Trajectory Optimization and Policy Learning for precise robot control
This work addresses the problem of efficient robot control for robotics researchers and practitioners, offering incremental improvements over prior hybrid approaches.
The paper tackles the challenge of combining reinforcement learning and trajectory optimization for robot control by proposing improvements like Sobolev learning and augmented Lagrangian techniques to learn global control policies faster, achieving quicker convergence compared to existing methods.
Reinforcement learning (RL) and trajectory optimization (TO) present strong complementary advantages. On one hand, RL approaches are able to learn global control policies directly from data, but generally require large sample sizes to properly converge towards feasible policies. On the other hand, TO methods are able to exploit gradient-based information extracted from simulators to quickly converge towards a locally optimal control trajectory which is only valid within the vicinity of the solution. Over the past decade, several approaches have aimed to adequately combine the two classes of methods in order to obtain the best of both worlds. Following on from this line of research, we propose several improvements on top of these approaches to learn global control policies quicker, notably by leveraging sensitivity information stemming from TO methods via Sobolev learning, and augmented Lagrangian techniques to enforce the consensus between TO and policy learning. We evaluate the benefits of these improvements on various classical tasks in robotics through comparison with existing approaches in the literature.