Pedestrian-Aware Motion Planning for Autonomous Driving in Complex Urban Scenarios

TL;DR

This work tackles autonomous vehicle motion planning in densely populated urban settings by introducing a risk-aware planner that couples a social force–based pedestrian simulator with a sampling-based trajectory planner. The framework evaluates harm and risk using $H(\xi)$ and $R(\xi)=p(\xi)\cdot H(\xi)$, where $p(\xi)$ is the collision probability derived from pedestrian and vehicle predictions, enabling a maximin safety criterion. A novel pedestrian simulation builds on the Social Force Model, incorporating sidewalks and crossings, offline policy computation via value iteration, and a 2 s vehicle-path prediction to generate realistic, interactive pedestrian behavior. The method is validated in 2D CommonRoad simulations, showing improved safety and maintained efficiency compared with baselines, and the authors release open-source code for further study, while acknowledging limitations such as real-time constraints and the need for real-world validation and stack integration.

Abstract

Motion planning in uncertain environments like complex urban areas is a key challenge for autonomous vehicles (AVs). The aim of our research is to investigate how AVs can navigate crowded, unpredictable scenarios with multiple pedestrians while maintaining a safe and efficient vehicle behavior. So far, most research has concentrated on static or deterministic traffic participant behavior. This paper introduces a novel algorithm for motion planning in crowded spaces by combining social force principles for simulating realistic pedestrian behavior with a risk-aware motion planner. We evaluate this new algorithm in a 2D simulation environment to rigorously assess AV-pedestrian interactions, demonstrating that our algorithm enables safe, efficient, and adaptive motion planning, particularly in highly crowded urban environments - a first in achieving this level of performance. This study has not taken into consideration real-time constraints and has been shown only in simulation so far. Further studies are needed to investigate the novel algorithm in a complete software stack for AVs on real cars to investigate the entire perception, planning and control pipeline in crowded scenarios. We release the code developed in this research as an open-source resource for further studies and development. It can be accessed at the following link: https://github.com/TUM-AVS/PedestrianAwareMotionPlanning

Figures (14)

• Figure 1: Example of a complex urban intersection illustrating the challenge for an AV to navigate safely and efficiently around a high density of pedestrians and human-driven vehicles
• Figure 2: Illustration of the motion planning problem. The ego vehicle plans its trajectory while considering dynamic obstacles and pedestrians. A sample collision constraint $\mathcal{C}_\text{coll}$, along with computed points of the selected trajectory and other non-selected trajectory candidates (orange), is shown.
• Figure 3: Illustration of the pedestrian simulation problem at the initial time step ($t=0$). Pedestrians are shown with their respective starting positions ($s_i$) and are grouped based on their shared target locations ($g_{1-3}$), with each group and respective target represented by a unique color. While pedestrians within the same group share a common goal, they act as independent agents. The illustration highlights key areas such as SWs and CWs and showcases behaviors like illegal road crossings, which occur when pedestrians are far from CWs, and their destination lies across the street.
• Figure 4: Overview of the simulation framework. The iterative process integrates a pedestrian simulator and a motion planner to evaluate safe and efficient vehicle trajectories in complex urban environments.
• Figure 5: Illustration of the forces influencing pedestrian motion in the scenario. The blue pedestrians are navigating toward their goal in the bottom-left corner, while the black pedestrians aim for their goal in the top-right corner, requiring them to cross the street. The gray dashed lines represent the vehicle predictions, which influence pedestrian behavior.