Reproduce “SanD-Planner: Sample-Efficient Diffusion Planner in B-Spline Space for Robust Local Navigation” 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

Jincheng Wang, Lingfan Bao, Tong Yang, Diego Martinez Plasencia, Jianhao Jiao, Dimitrios Kanoulas

UCL researchers developed SanD-Planner, a diffusion-based local navigation planner leveraging B-spline trajectory representation. It achieves robust local navigation and zero-shot sim-to-real transfer with significantly reduced training data, performing comparably or better than prior methods that use far more demonstrations.

Key insights

  • Embedding structural geometric priors, such as B-spline trajectory parameterization, into learning-based planners dramatically improves sample efficiency, robustness, and trajectory quality.
  • Decoupling complex trajectory generation from explicit safety assessment with a geometric critic simplifies the learning problem and enhances reliability and interpretability.
  • Augmenting the conditioning signal of generative models with temporal context, like previous velocity, is crucial for achieving consistent and stable behaviors in multi-modal navigation scenarios.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.