The Impact of Class Uncertainty Propagation in Perception-Based Motion Planning

TL;DR

This work compares two novel prediction-planning pipelines with varying levels of uncertainty propagation on the recently-released nuPlan planning benchmark and finds that the method incorporating upstream uncertainty propagation demonstrates superior generalization to complex closed-loop scenarios.

Abstract

Autonomous vehicles (AVs) are being increasingly deployed in urban environments. In order to operate safely and reliably, AVs need to account for the inherent uncertainty associated with perceiving the world through sensor data and incorporate that into their decision-making process. Uncertainty-aware planners have recently been developed to account for upstream perception and prediction uncertainty. However, such planners may be sensitive to prediction uncertainty miscalibration, the magnitude of which has not yet been characterized. Towards this end, we perform a detailed analysis on the impact that perceptual uncertainty propagation and calibration has on perception-based motion planning. We do so by comparing two novel prediction-planning pipelines with varying levels of uncertainty propagation on the recently-released nuPlan planning benchmark. We study the impact of upstream uncertainty calibration using closed-loop evaluation on the nuPlan challenge scenarios. We find that the method incorporating upstream uncertainty propagation demonstrates superior generalization to complex closed-loop scenarios.

Figures (3)

• Figure 1: A direct comparison of prediction-planning performance between our novel uncertainty-aware stochastic MPC (UA-SMPC) when paired with Trajectron++ SalzmannIvanovicEtAl2020 (left), an uncertainty-agnostic prediction model, and HAICU IvanovicLeeEtAl2022 (right), an uncertainty-aware prediction model. HAICU produces significantly more calibrated trajectory predictions than Trajectron++. As a result, HAICU-backed UA-SMPC generates smoother, lane-respecting motion plans, while Trajectron++-backed UA-SMPC swerves aggressively to avoid collisions with the poorly-calibrated predictions. Ego vehicle is shown in orange, nearby non-ego vehicles in green, and distant non-ego vehicles in grey with pink tracks.
• Figure 2: In both Trajectron++ and HAICU, the scene is modeled as a graph with nodes representing agents and edges representing their interactions. However, one key distinction is that HAICU incorporates both the state vector and class probabilities into its node and edge histories, whereas Trajectron++ only includes the state vector.
• Figure 3: A rollout comparison of UA-SMPC paired with Trajectron++ SalzmannIvanovicEtAl2020 (top) and HAICU IvanovicLeeEtAl2022 (bottom) over three sequential timesteps in an open-loop right turn scenario. Trajectron++ fails to predict the trailing agent’s right turn, while HAICU accurately forecasts it with increasing likelihood, leading to a safer motion plan. See \ref{['fig:teaser']} for legend.