Risk-Aware Navigation for Mobile Robots in Unknown 3D Environments

TL;DR

This work extends the Lambda-Field framework from 2D to 3D to enable risk-aware navigation in unknown environments using 3D lidar-derived DEMs. It introduces a physically interpretable risk measure based on the maximum energy absorbed by wheels during collisions, and formulates a hard-risk constraint NMPC for trajectory optimization under Ackermann dynamics. The approach permits crossing small traversable obstacles (e.g., speed bumps) by quantifying and controlling risk, validated through simulations in urban-like scenarios with real perception data. This method strengthens autonomous mobile robots' capability to operate safely and efficiently in complex 3D environments where pure collision avoidance is insufficient.

Abstract

Autonomous navigation in unknown 3D environments is a key issue for intelligent transportation, while still being an open problem. Conventionally, navigation risk has been focused on mitigating collisions with obstacles, neglecting the varying degrees of harm that collisions can cause. In this context, we propose a new risk-aware navigation framework, whose purpose is to directly handle interactions with the environment, including those involving minor collisions. We introduce a physically interpretable risk function that quantifies the maximum potential energy that the robot wheels absorb as a result of a collision. By considering this physical risk in navigation, our approach significantly broadens the spectrum of situations that the robot can undertake, such as speed bumps or small road curbs. Using this framework, we are able to plan safe trajectories that not only ensure safety but also actively address the risks arising from interactions with the environment.

Figures (4)

• Figure 1: Example of a situation that a vehicle might encounter while navigating. Speed bumps are frequent obstacles an intelligent vehicle has to safely overcome. With our framework, the vehicle is able to make more effective decisions based on the level of risk it is allowed to take.
• Figure 2: Example of Lambda-Field of the environment depicted in \ref{['fig:intro']}. The area containing the speed bump and the cones is enlarged and shown at different times. The lidar range (vertical line) and the robot (box) pose for each of these times are illustrated in teal ($t_1$), dull green ($t_2$) and blue ($t_3$). The left curb (light blue) and the traffic lights (green) seen in \ref{['fig:intro']} are outlined in dashed lines.
• Figure 3: Modeling of the wheel of radius $R$ in collision with the curb (in blue) of height $H$, with a speed $v$ and angle $\Psi$. The deformation of the wheel due to the collision is approximated with the deformation of a spring of stiffness $k_r$.
• Figure 4: Example of an environment where an unexpected obstacle (speed bump) occurred on the reference path (dashed green). We investigated three risk thresholds, 0 (left), 3 (middle) and 40 (right). For each threshold, the path taken by the robot is in blue. Its velocity and risk are depicted in purple and red. Events: (1): Speed bump came into the robot's perception field. (2): Robot started to climb the speed bump. (3): Robot reached the speed bump top. (4): Robot started to get down. (5): Robot reached the asphalt. The gray shaded areas show the duration when the robot traversed the speed bump.