Accurate motion prediction is critical for safe autonomous driving, yet even strong forecasters leave residual errors in their multimodal proposals. The paper introduces FutureLift, a method that lifts trajectory proposals into bidirectional future graphs to address these remaining inaccuracies.

Existing refinement approaches either compress complete proposals or update their segments individually. FutureLift instead constructs a graph that captures relationships across the future trajectory, allowing for more coherent refinements than segment-wise or compression-based methods.

The work is presented on arXiv as a new direction for motion forecasting. While the abstract does not include quantitative results, the proposed graph-based lifting mechanism offers a distinct alternative to current refinement strategies.