Reproduce “SG-LSTM: Social Group LSTM for Robot Navigation Through Dense Crowds” 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
Rashmi Bhaskara, Maurice Chiu, Aniket Bera
Purdue University researchers developed SG-LSTM, a deep learning model that significantly improves robot navigation in dense crowds by explicitly modeling social group dynamics. It achieves lower displacement errors in pedestrian trajectory prediction and reduces computation time by over 50% compared to previous individual-centric approaches, facilitating more socially acceptable and efficient robot operation.
Key insights
- Explicitly modeling human social groups as cohesive entities, instead of only individuals, improves pedestrian trajectory prediction accuracy and computational efficiency in dense crowds.
- A hierarchical LSTM architecture with a group-level social pooling layer can effectively capture complex group dynamics while reducing the number of parameters needed for scene representation.
- Leveraging depth information for spatial coordinate calculation and integrating group predictions with a Generalized Velocity Obstacles (GVO) approach enables socially compliant robot navigation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.