EIDA: Execution-Interface Dynamics Adaptation for Real-to-Sim-to-Real Robot Navigation

2026-10-01

AI Summary

This study addresses navigation failures in sim-to-real transfer caused by motion and feedback discrepancies induced by velocity commands. We propose a dynamic adaptation method for execution interfaces that decouples pose increment modeling from velocity feedback. By fitting target platform response data to update simulator geometry and policy inputs, the approach enables real-time adaptation without reconstructing actuator dynamics. Combined with GPU-parallelized lightweight simulation, historical-data-driven feedback prediction, and a pose increment update algorithm, it significantly enhances transfer efficiency. Experiments demonstrate state-of-the-art success rates on standard benchmarks. Furthermore, real-world deployment on a Go2 quadruped robot achieves a 20/20 collision-free rate in static scenarios, substantially outperforming the baseline (4/20) and validating the effectiveness of the proposed method.

Abstract

Simulation-to-robot transfer can fail when velocity commands produce motion and feedback that differ from those modeled during policy training. We present execution-interface dynamics adaptation (EIDA), which fits these responses from target-platform execution data without reconstructing actuator dynamics. A model of body-frame pose increments updates simulator geometry, while a separate model predicts the velocity feedback observed by the policy; a short history of velocity feedback is included in the policy input. The fitted models are used within a lightweight GPU-parallel simulator. On the full Jackal and Go2 validation sets, the fitted models reduced position and yaw prediction errors relative to the simulator's predefined motion model. Across 100 benchmark navigation environments evaluated in a separate physics-based simulator, EIDA achieved the highest success rate and navigation score among the compared learned policies, both with and without global guidance. Feedback ablations further supported the need to match policy-facing velocity estimates. On a physical Unitree Go2, EIDA reached the goal without collision in all 20 static-scene trials, compared with 4 of 20 for the baseline. These results show that execution-interface adaptation can improve navigation transfer without detailed actuator simulation.

Problem

• Sim-to-Real transfer
• Robot navigation
• Dynamics mismatch
• Execution-interface adaptation

Innovation

• Execution-Interface Dynamics Adaptation
• Sim-to-Real Transfer
• Robot Navigation
• GPU-Parallel Simulator
• Velocity Feedback