Reproduce “EmoBipedNav: Emotion-aware Social Navigation for Bipedal Robots with Deep Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wei Zhu, Abirath Raju, Abdulaziz Shamsah, Anqi Wu, Seth Hutchinson, Ye Zhao
EmoBipedNav trains a deep-reinforcement-learning policy for social navigation using recent LiDAR grid maps marked with collision and emotion-dependent discomfort zones. It outputs center-of-mass speed and torso heading, with simulated Digit’s full-body dynamics and controller in the training loop. In 500 tests, it achieved 0.950 success and 226 discomfort steps, versus OGM’s 0.944 and 312; it was not fastest.
Key insights
- Direct DRL training with full-body bipedal robot dynamics effectively mitigates performance degradation and tracking errors that arise from discrepancies between reduced-order models and real-world robot locomotion.
- Integrating real-time pedestrian emotional cues through variable discomfort zones enables bipedal robots to exhibit more socially acceptable and courteous navigation behaviors in human-populated spaces.
- Sequential LiDAR Grid Maps (LGMs) provide a robust and information-rich observation space, superior to traditional occupancy grid maps, for capturing dynamic social interactions and spatio-temporal environmental evolution for bipedal social navigation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.