Control Transformer: Robot Navigation in Unknown Environments through PRM-Guided Return-Conditioned Sequence Modeling

TL;DR

This paper introduces Control Transformer, a novel approach combining sequence modeling and sampling-based planning to enable robots to navigate long-horizon, partially-observed environments and transfer to real-world scenarios without additional training.

Contribution

The paper presents a new framework that integrates PRM-guided sequence modeling for long-horizon robot navigation in unknown environments, demonstrating effective sim2real transfer.

Findings

Successfully navigates mazes with MuJoCo robots

Transfers to real Turtlebot3 robot in zero-shot setting

Outperforms baseline methods in long-horizon tasks

Abstract

Learning long-horizon tasks such as navigation has presented difficult challenges for successfully applying reinforcement learning to robotics. From another perspective, under known environments, sampling-based planning can robustly find collision-free paths in environments without learning. In this work, we propose Control Transformer that models return-conditioned sequences from low-level policies guided by a sampling-based Probabilistic Roadmap (PRM) planner. We demonstrate that our framework can solve long-horizon navigation tasks using only local information. We evaluate our approach on partially-observed maze navigation with MuJoCo robots, including Ant, Point, and Humanoid. We show that Control Transformer can successfully navigate through mazes and transfer to unknown environments. Additionally, we apply our method to a differential drive robot (Turtlebot3) and show zero-shot sim2real transfer under noisy observations.