Estimating Visibility from Alternate Perspectives for Motion Planning with Occlusions

TL;DR

The paper tackles visibility under occlusions in autonomous vehicle planning by introducing the Alternate Perspective Cost Map (APCM), a grid-based representation that quantifies how much of occluded regions would be observed from candidate locations. APCM is computed by identifying reachable occluded space $\mathcal{U}_t^r$, selecting observation locations $\mathcal{S}_t$, and estimating perspective probabilities via ray-casting against an occupancy map, then aggregating into $\mathbf{P}_t$. The APCM is integrated into an MPPI-based MPC to produce trajectories that reveal occluded regions earlier and robustly handle multiple occlusions, with a GPU-accelerated implementation that meets real-time requirements. Simulation in CARLA demonstrates improved performance in dense clutter compared to baselines such as Higgins and Andersen, and the method is generalizable to dynamic occupancy extensions and other robot domains. The work provides a practical, scalable approach to perception-aware planning in occluded environments.

Abstract

Visibility is a crucial aspect of planning and control of autonomous vehicles (AV), particularly when navigating environments with occlusions. However, when an AV follows a trajectory with multiple occlusions, existing methods evaluate each occlusion individually, calculate a visibility cost for each, and rely on the planner to minimize the overall cost. This can result in conflicting priorities for the planner, as individual occlusion costs may appear to be in opposition. We solve this problem by creating an alternate perspective cost map that allows for an aggregate view of the occlusions in the environment. The value of each cell on the cost map is a measure of the amount of visual information that the vehicle can gain about the environment by visiting that location. Our proposed method identifies observation locations and occlusion targets drawn from both map data and sensor data. We show how to estimate an alternate perspective for each observation location and then combine all estimates into a single alternate perspective cost map for motion planning.

Figures (7)

• Figure 1: An example of two possible trajectories for an autonomous vehicle with occlusions highlighted. In (\ref{['fig:visibility-demo-occluded']}), the AV maintains the nominal trajectory resulting in an occlusion that hides two pedestrians (darker shadow in upper right). In (\ref{['fig:visibility-demo-revealed']}), the AV takes a position on the left side, providing earlier perception of the crosswalk entry.
• Figure 2: Construction of the Perspective Cost Map.
• Figure 3: The observation locations are highlighted along the intended trajectory. Phantom pedestrians are inserted into the occluded space in front of the truck from where they could enter the roadway.
• Figure 4: The magnitude of the response to an occlusion is controlled by the weighting parameter $M or \lambda$. We placed a single vehicle and selected the weight values that resulted in similar responses, creating a basis for equitable comparisons in dense clutter scenarios.
• Figure 5: Experiments are run in one of these four scenarios, each with randomly placed parked cars. The first two are sparsely cluttered: (\ref{['fig:scenario-straight']}) a multilane straight section of road, and (\ref{['fig:scenario-intersection']}) an open intersection. The second two are more densely cluttered: (\ref{['fig:scenario-curve']})a slow curve, and (\ref{['fig:scenario-curve']}) a two-lane road outside a park. A light blue line indicates the nominal path.

Theorems & Definitions (5)

• Remark 2.1: Extension to Dynamic Occupancy
• Remark 2.2: Restricted to 2D
• Remark 3.1: Importance of Thresholding
• Remark 3.2: Proportion of $\mathcal{U}_t^r$ observed
• Remark 6.1: Deviation of None Method