End-to-End Training of Deep Visuomotor Policies

TLDR

Policy search enables robots to learn control policies, yet real‑world use often depends on hand‑engineered perception, state estimation, and low‑level control modules. This work investigates whether jointly training perception and control end‑to‑end yields superior performance compared to training each component separately. We propose a deep CNN policy that maps raw images to motor torques, trained with partially observed guided policy search that turns policy search into supervised learning using trajectory‑centric reinforcement learning, and evaluate it on manipulation tasks such as screwing a cap onto a bottle against prior methods.

Abstract

Policy search methods can allow robots to learn control policies for a wide range of tasks, but practical applications of policy search often require hand-engineered components for perception, state estimation, and low-level control. In this paper, we aim to answer the following question: does training the perception and control systems jointly end-to-end provide better performance than training each component separately? To this end, we develop a method that can be used to learn policies that map raw image observations directly to torques at the robot's motors. The policies are represented by deep convolutional neural networks (CNNs) with 92,000 parameters, and are trained using a partially observed guided policy search method, which transforms policy search into supervised learning, with supervision provided by a simple trajectory-centric reinforcement learning method. We evaluate our method on a range of real-world manipulation tasks that require close coordination between vision and control, such as screwing a cap onto a bottle, and present simulated comparisons to a range of prior policy search methods.