Efficient Trajectory Forecasting and Generation with Conditional Flow Matching

TL;DR

Trajectory Conditional Flow Matching (T-CFM) is introduced, a novel approach using flow matching techniques to learn a solver time-varying vector field for efficient, fast trajectory generation, enabling real-time decision making in robotics.

Abstract

Trajectory prediction and generation are crucial for autonomous robots in dynamic environments. While prior research has typically focused on either prediction or generation, our approach unifies these tasks to provide a versatile framework and achieve state-of-the-art performance. While diffusion models excel in trajectory generation, their iterative sampling process is computationally intensive, hindering robotic systems' dynamic capabilities. We introduce Trajectory Conditional Flow Matching (T-CFM), a novel approach using flow matching techniques to learn a solver time-varying vector field for efficient, fast trajectory generation. T-CFM demonstrates effectiveness in adversarial tracking, real-world aircraft trajectory forecasting, and long-horizon planning, outperforming state-of-the-art baselines with 35% higher predictive accuracy and 142% improved planning performance. Crucially, T-CFM achieves up to 100$\times$ speed-up compared to diffusion models without sacrificing accuracy, enabling real-time decision making in robotics. Codebase: https://github.com/CORE-Robotics-Lab/TCFM

Figures (6)

• Figure 1: Trajectory Conditional Flow Matching (T-CFM) is our novel trajectory prediction and generation framework. The model is capable of generating trajectories in a single step, outperforming prior generative modeling work by learning a time-varying vector field to sample trajectories.
• Figure 2: Overview of Trajectory Conditional Flow Matching. The flow matching formulation defines intermediate trajectories as a linear combination between the prior noise distribution ($\tau_0$) and data distribution ($\tau_1)$. The sampling procedure then utilizes the learned flow field generated by the model to create samples.
• Figure 3: Trajectory Forecasting and Planning Domains: Our T-CFM framework is applicable many trajectory modeling tasks, with Adversarial Tracking, Trajectory Forecasting, and Long-Horizon Planning domains shown here.
• Figure 4: Comparison our model with the diffusion-based CADENCE model. Our method achieves better ADE on the entire prediction horizon (left) while also maintaining performance when the number of sampling steps is lowered (right).
• Figure 5: Visualization of Sampling Procedure between Diffuser (top) and T-CFM (bottom) in Maze2D-Medium. In just two sampling steps, we show that T-CFM can successfully plan a path between the start and end unlike Diffuser.