Latent Imagination Facilitates Zero-Shot Transfer in Autonomous Racing
This addresses the challenge of sim2real transfer in robotics, demonstrating improved generalization for autonomous racing, though it is incremental as it extends existing world model methods to a new application domain.
The paper tackled the problem of applying world models for sample-efficient deep reinforcement learning to real-world autonomous vehicle control, specifically in F1TENTH racing robots with high-dimensional LiDAR sensors, showing that model-based agents outperform model-free agents in performance, sample efficiency, task completion, and generalization across increasingly complex tracks.
World models learn behaviors in a latent imagination space to enhance the sample-efficiency of deep reinforcement learning (RL) algorithms. While learning world models for high-dimensional observations (e.g., pixel inputs) has become practicable on standard RL benchmarks and some games, their effectiveness in real-world robotics applications has not been explored. In this paper, we investigate how such agents generalize to real-world autonomous vehicle control tasks, where advanced model-free deep RL algorithms fail. In particular, we set up a series of time-lap tasks for an F1TENTH racing robot, equipped with a high-dimensional LiDAR sensor, on a set of test tracks with a gradual increase in their complexity. In this continuous-control setting, we show that model-based agents capable of learning in imagination substantially outperform model-free agents with respect to performance, sample efficiency, successful task completion, and generalization. Moreover, we show that the generalization ability of model-based agents strongly depends on the choice of their observation model. We provide extensive empirical evidence for the effectiveness of world models provided with long enough memory horizons in sim2real tasks.