PrismAD: Decoupled Planning via Semantic Mixture-of-Planners for End-to-End Autonomous Driving
2026-07-11 • Robotics
RoboticsComputer Vision and Pattern Recognition
AI summaryⓘ
The authors propose PrismAD, a new approach for self-driving cars that breaks down the driving environment into three parts: how vehicles interact, the shape of the road, and the driver's intent. Instead of trying to handle everything together, their system has separate 'experts' each focusing on one part, which helps each expert learn better. A smart router then combines these experts' suggestions to plan the car's next move. Their tests show that PrismAD performs well compared to other methods.
autonomous drivingmotion planningsemantic mixture-of-plannersscene tokensagent interactionroad geometrydriving intentionroutingnuScenes datasetNeuroNCAP benchmark
Authors
Kang Ding, Zhigui Lin, Hongsong Wang, Jie Gui, Qi Liu, Zhe Wang, Luqi Tang, Lei He
Abstract
This letter presents PrismAD, a decoupled end-to-end autonomous driving framework based on a Semantic Mixture-of-Planners. Existing planners usually aggregate heterogeneous scene tokens into a coupled representation space, forcing a single planning branch to jointly model agent interaction, road geometry, and driving intention. Such coupling may weaken factor-specific reasoning and obscure the contribution of different planning cues. To address this limitation, PrismAD partitions scene tokens into interaction, geometry, and intent groups, and assigns them to independent planning experts with the same architecture but separate parameters. Each expert learns a specialized motion-planning representation, while a semantics-aware router adaptively aggregates expert predictions with separate routing weights for motion prediction and ego planning. Sparse top-$K$ activation with noisy gating is further introduced to improve routing robustness and reduce unnecessary expert computation. Extensive experiments on the nuScenes open-loop dataset and NeuroNCAP closed-loop benchmark demonstrate that PrismAD exhibits competitive performance. Our code will be released soon.