Reproduce “Efficient Trajectory Forecasting and Generation with Conditional Flow Matching” 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
Sean Ye, Matthew Gombolay
T-CFM generates conditional trajectories for forecasting and planning by learning a time-dependent direction field that moves noisy candidates toward data-like trajectories, rather than iteratively denoising them. In Maze2D, it scores 109.9 versus Diffuser’s 34.2 at one sampling step, while nearly tying at 256 steps; however, Diffuser performs better in the single-task Large maze at 256 steps.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.