Towards Safe Autonomous Driving: A Real-Time Motion Planning Algorithm on Embedded Hardware

TL;DR

This work tackles fail-operational safety in autonomous driving by proposing an embedded fallback planner that runs on automotive-grade hardware with an RTOS to provide active safety in case the main planner fails. It develops a real-time, sampling-based motion planner operating in curvilinear coordinates $(s,d)$ anchored to a non-self-intersecting reference path $\Gamma$, with candidate trajectories generated by end-states $g_k(\tau)$ and polynomials $s(t)=a_0+...+a_4 t^4$, $d(t)=b_0+...+b_5 t^5$, and evaluated under dynamic and geometric constraints through a weighted cost $J_{\text{sum}}$. The system communicates with the high-level ROS2 stack via a minimal DDS bridge on Zephyr, enabling a safety island that outputs $\xi^{\star}$ for the controller. Experimental results on an automotive-grade board show deterministic timing with bounded latency and minimal jitter, validating feasibility as a backup and informing future integration into a Minimal Risk Maneuver (MRM) module for a certified fail-operational safety framework.

Abstract

Ensuring the functional safety of Autonomous Vehicles (AVs) requires motion planning modules that not only operate within strict real-time constraints but also maintain controllability in case of system faults. Existing safeguarding concepts, such as Online Verification (OV), provide safety layers that detect infeasible planning outputs. However, they lack an active mechanism to ensure safe operation in the event that the main planner fails. This paper presents a first step toward an active safety extension for fail-operational Autonomous Driving (AD). We deploy a lightweight sampling-based trajectory planner on an automotive-grade, embedded platform running a Real-Time Operating System (RTOS). The planner continuously computes trajectories under constrained computational resources, forming the foundation for future emergency planning architectures. Experimental results demonstrate deterministic timing behavior with bounded latency and minimal jitter, validating the feasibility of trajectory planning on safety-certifiable hardware. The study highlights both the potential and the remaining challenges of integrating active fallback mechanisms as an integral part of next-generation safeguarding frameworks. The code is available at: https://github.com/TUM-AVS/real-time-motion-planning

Figures (8)

• Figure 1: System architecture illustrating the integration of a safety island that hosts both a Supervisor and an embedded planner. The concept enables safety monitoring and fallback trajectory generation independently of the main software stack, forming the foundation for fail-operational .
• Figure 2: Overview of the proposed communication architecture between the and the Embedded System. The runs a simulation, while the executes our motion planner. A lightweight DDS bridge enables data exchange across both platforms.
• Figure 3: Overview of the planning workflow on the embedded platform. The initialization phase (left) is executed once at startup, followed by the cyclic motion planning loop (right).
• Figure 4: Qualitative simulation result in the CommonRoad environment, showing a left-turn scenario at an unsignalized intersection. The generated candidate trajectories follow the reference path, illustrating how the planner maintains smooth motion during the intended turning maneuver.
• Figure 5: Generation and evaluation of trajectory samples by the planner along $\Gamma$. Valid trajectories are color-coded based on their cost, ranging from green (low cost) to red (high cost). The optimal trajectory is shown in black.