Motion forecasting for autonomous driving in a streaming world

Predicting the future and making decisions accordingly is essential for the safety of autonomous driving; thus, trajectory forecasting becomes a widely-studied problem for autonomous navigation by predicting the movements of traffic participants. Although the forecasting algorithms have advanced rapidly in recent years, we discover that the formulation of existing benchmarks has gaps with the real world and causes systematic failures of motion forecasting systems. Specifically, they evaluate forecasting based on independent snapshots of trajectories, which are not representative of real-world applications that operate on a continuous stream of data. To bridge this gap, this thesis begins by introducing a benchmark that champions the streaming formulation of real-world traffic for motion forecasting. Its most essential design is to continuously query future trajectories of other agents on streaming data, and we refer to it as "streaming forecasting." Our benchmark inherently captures the disappearance and re-appearance of agents, presenting the emergent challenge of forecasting for occluded agents, which is a safety-critical problem yet overlooked by snapshot-based benchmarks in autonomous driving. Moreover, forecasting in the context of continuous timestamps naturally asks for temporal coherence between predictions from adjacent timestamps. Based on this benchmark, this thesis further provides solutions and analysis for streaming forecasting to improve the safety of autonomous driving. We propose a plug-and-play meta-algorithm called "Predictive Streamer" that can adapt any snapshot-based forecaster into a streaming forecaster. Our algorithm estimates the states of occluded agents by propagating their positions with multi-modal trajectories containing multiple possible future states and leverages differentiable filters to ensure temporal consistency. Both occlusion reasoning and temporal coherence strategies significantly improve forecasting quality, resulting in 25% smaller endpoint errors for occluded agents and 10-20% smaller fluctuations of trajectories. Finally, this thesis is intended to generate interest within the community by highlighting the importance of addressing motion forecasting in its intrinsic streaming setting. Given the recent trend of end-to-end autonomous driving research, this thesis also contributes to developing forecasting algorithms and metrics under a streaming perception perspective.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Pang, Ziqi

Contributors dc:contributor

• Wang, Yuxiong

• Autonomous Driving
• Motion Forecasting
• Computer Vision

Statement dc:rights

• Copyright 2024 Ziqi Pang

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/124574

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F124574