Predictive Training with Latent Imagination for Visual Quadruped Navigation
2026-07-20 • Robotics
RoboticsArtificial Intelligence
AI summaryⓘ
The authors study how legged robots navigate around moving obstacles using a method that helps them predict what will happen next instead of just reacting to the current scene. They trained their robot's navigation system to anticipate future obstacle movements by adding a special predictor during training but then removed it when the robot actually moves, so it doesn’t slow down performance. Tests in simulation and with a real robot showed this approach makes navigation safer and more successful, without requiring extra computation during movement. Their system also worked well in new environments without extra tuning.
reinforcement learninglegged robotsnavigation policyrecurrent neural networksLSTMpredictive supervisiondynamic environmentssim-to-real transfercollision avoidanceauxiliary training
Authors
Yancheng Zhu, Wanli Ma, Chen Han, Irvin Haozhe Zhan, Bingfeng Qin, Yixin Xu
Abstract
Reinforcement-learning navigation policies for legged robots select actions reactively from current observations and short-term memory, with limited capacity to anticipate how moving obstacles will evolve in the near future. In dynamic environments, this reactivity causes the robot to respond too late because collision risk depends on short-horizon scene structure rather than on current obstacle positions alone. Lightweight predictive supervision applied to the policy's recurrent state during training can encode anticipatory obstacle dynamics without modifying the inference-time controller. We augment a reactive LSTM-SRU navigation backbone with an auxiliary JEPA-style predictor and SIGReg regularization: during training, the predictor supervises the deterministic hidden state to anticipate its own next state; at inference, it is fully discarded, incurring zero additional computational cost. On simulated and real-world navigation benchmarks with dynamic obstacles, our method substantially improves navigation success while reducing collision rates through the predictive training signal alone, without additional inference-time parameters. Real-robot deployment on a Unitree Go2 demonstrates zero-shot sim-to-real transfer: the controller navigates cluttered indoor and dynamic outdoor environments without fine-tuning, with evasive behavior consistent with the collision reduction observed in simulation.