Map representations for vehicle trajectory forecasting.

Trajectory forecasting (and planning) techniques require a good understanding of the world around them to predict their next steps.

To this end, we usually have to rely on previously acquired and curated HDMaps that contain all the information a model might need. Unfortunately, HDMaps are largely made of somewhat symbolic information (this map element is here, it is linked to another, and so on).

I am investigating how this information should be encoded to provide the best information to forecasting systems. In particular, I am interested in how static knowledge (e.g. the behavior on a well known street) can be made understandable to the network.

Furthermore, I am intersted in how HDMap acquisition and maintenance can be partially automated by combining existing maps and sensor data.

Work on the matter:

• Supervising the work of Li Yang on HDMap editing for data augmentation.
• Preprint paper “Mind the map! Accounting for existing map information when estimating online HDMaps from sensor data”. In this work, we propose a way to incorporate information on existing maps in online HDMap estimation and show this leads drastic improvements of performance.
• Workshop paper “Exploring the Road Graph in trajectory forecasting” in the SG2RL workshop at ICCV 2023. This is a preliminary study on the effect of various interventions on the road graph representation with regards to forecasting performance.