Agile robot navigation improved by teaching from expert guidance
TAME-Nav: Temporal Asymmetric Mixture-of-Experts Distillation for Agile and Versatile Robot Navigation
Robotics
Summary
Robots need to move smoothly and safely through complicated and changing spaces, but one single way to decide how to move often fails in different situations. The authors created a system where a smart teacher robot uses clear and detailed information to guide a student robot that only sees noisy or partial data. This teaching method helps the student robot learn better ways to navigate, even in unexpected or tricky scenarios. The new system was tested both in simulations and in the real world on a four-legged robot, showing fewer collisions and better success at avoiding obstacles.
What this means in practice
- •For robotics engineers: Enable legged robots to navigate challenging dynamic environments more reliably using learned expert guidance with corrupted sensory input.
- •For autonomous mobile robot developers: Improve simulation-to-real robot navigation by training with asymmetric domain randomization and expert-teacher methods for varied or unforeseen scenarios.
Authors
Lingjie Zhang, Jun Xue, Shanze Wang, Haoyu Xi, Xinming Zhang, Xiaoyu Shen, Wei Zhang
Abstract
Agile and reliable navigation is a core capability for embodied robots navigating complex spatial constraints and dynamic interactions. However, a single policy often struggles to accommodate the distinct decision-making demands of diverse scenarios. We present TAME-Nav, an asymmetric Teacher-Student navigation framework built around a temporal soft mixture of experts (Soft-MoE). The Teacher combines clean observations with causally maintained dynamic-obstacle tracks to construct privileged representations, while the Student learns complementary navigation features from corrupted observation histories through temporal encoding, soft expert routing, and latent distillation. Procedural scenario generation and parallel task-level simulation broaden training coverage, while deployment-informed asymmetric domain randomization supports sim-to-real transfer. Across in-distribution and unseen compositional scenarios, TAME-Nav achieves higher success rates, lower collision rates, and higher navigation scores than representative baselines. On a real quadruped robot, it completes all twenty trials across local-trap escape and narrow-passage navigation with sudden dynamic obstacles, with only one minor contact.