Reproduce “Stochastic Occupancy Grid Map Prediction in Dynamic Scenes” 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
Zhanteng Xie, Philip Dames
SOGMP and SOGMP++ predict multiple plausible future occupancy-grid maps for moving robots by combining ego-motion compensation, ConvLSTM scene modeling, and VAE-based stochastic generation. Across simulated and real robot datasets, they improve WMSE, SSIM, and OSPA over image-prediction baselines, while the uncertainty-aware DWA-SOGMP-PU planner raises simulated navigation success to 0.89 in scenes with 35 pedestrians and 0.82 with 45 pedestrians.
Key insights
- Explicit ego-motion compensation is central: removing it worsens WMSE, SSIM, and OSPA across simulated and real test datasets.
- The stochastic VAE produces diverse future configurations, with estimated entropy stabilizing as more samples are drawn and increasing in scenes containing more objects.
- Static-map information mainly improves preservation of scene structure and long-horizon SSIM, while dynamic-object localization remains similar between SOGMP and SOGMP++.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.