Automatic Environment Shaping is the Next Frontier in RL
This addresses the problem of high human effort in setting up RL tasks for robotics, which is incremental as it builds on existing sim-to-real RL methods.
The paper argues that the main bottleneck in scaling reinforcement learning (RL) to diverse robotic tasks is the manual effort required to shape training environments, such as designing observations, actions, rewards, and dynamics, rather than algorithmic improvements. It calls for the RL community to focus on automating environment shaping procedures to enable robots to learn tasks autonomously.
Many roboticists dream of presenting a robot with a task in the evening and returning the next morning to find the robot capable of solving the task. What is preventing us from achieving this? Sim-to-real reinforcement learning (RL) has achieved impressive performance on challenging robotics tasks, but requires substantial human effort to set up the task in a way that is amenable to RL. It's our position that algorithmic improvements in policy optimization and other ideas should be guided towards resolving the primary bottleneck of shaping the training environment, i.e., designing observations, actions, rewards and simulation dynamics. Most practitioners don't tune the RL algorithm, but other environment parameters to obtain a desirable controller. We posit that scaling RL to diverse robotic tasks will only be achieved if the community focuses on automating environment shaping procedures.