Adaptive-WAM: Quality-Guided Early-Exit Planning from Intermediate Video-Diffusion Features
Adaptive-WAM: Quality-Guided Early-Exit Planning from Intermediate Video-Diffusion Features
This paper presents Adaptive-WAM, a planning framework built on a Wan2.2-5B video diffusion backbone that uses intermediate-layer features to decode trajectories and terminates inference early when quality thresholds are met. The method avoids iterative future-video synthesis, achieving 90.8 PDMS on NAVSIM with adaptive routing, 89.9 EPDMS on NAVSIM v2, and transferring to nuScenes with 0.88 m average L2 error and a 0.08% collision rate without fine-tuning. On an A100 GPU, the adaptive planner averages 170 ms end-to-end latency, improving over fixed-depth baselines while maintaining strong performance. Source code will be released.
The study introduces Adaptive-WAM, a quality-guided early-exit planning method that dynamically allocates computation in a video diffusion backbone to achieve competitive performance on NAVSIM and transfer to nuScenes without domain-specific fine-tuning.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
PDMS score
90.8
View original evidence
ng to trajectory quality. On NAVSIM, the adaptive single-trajectory planner achieves 90.8 PDMS; a separate fixed-exit variant reaches 92.6 PDMS with 64 proposals. It further obtains 89.9 EPDMS on NAVSIM v2, yielding the best reported results among the compared front-view video world-model planners. Without target-domain fine-tuning, Adaptive-WAM transfers to nuScenes with 0.88 m average L2 error and a 0.08\% collisio
Open source S5PDMS score
92.6
View original evidence
ng to trajectory quality. On NAVSIM, the adaptive single-trajectory planner achieves 90.8 PDMS; a separate fixed-exit variant reaches 92.6 PDMS with 64 proposals. It further obtains 89.9 EPDMS on NAVSIM v2, yielding the best reported results among the compared front-view video world-model planners. Without target-domain fine-tuning, Adaptive-WAM transfers to nuScenes with 0.88 m average L2 error and a 0.08\% collisio
Open source S5EPDMS score
89.9
View original evidence
ng to trajectory quality. On NAVSIM, the adaptive single-trajectory planner achieves 90.8 PDMS; a separate fixed-exit variant reaches 92.6 PDMS with 64 proposals. It further obtains 89.9 EPDMS on NAVSIM v2, yielding the best reported results among the compared front-view video world-model planners. Without target-domain fine-tuning, Adaptive-WAM transfers to nuScenes with 0.88 m average L2 error and a 0.08\% collisio
Open source S5average L2 error
0.88
View original evidence
ng to trajectory quality. On NAVSIM, the adaptive single-trajectory planner achieves 90.8 PDMS; a separate fixed-exit variant reaches 92.6 PDMS with 64 proposals. It further obtains 89.9 EPDMS on NAVSIM v2, yielding the best reported results among the compared front-view video world-model planners. Without target-domain fine-tuning, Adaptive-WAM transfers to nuScenes with 0.88 m average L2 error and a 0.08\% collisio
Open source S5planning latency
170
View original evidence
n rate. On an A100, adaptive routing improves PDMS from 90.62 to 90.79 while averaging 170 ms end-to-end planning latency, approximately 10\% below the 190 ms fixed block-15 planner and 47\% below the 320 ms fixed full-depth planner. Code will be released.
Open source S6planning latency
190
View original evidence
n rate. On an A100, adaptive routing improves PDMS from 90.62 to 90.79 while averaging 170 ms end-to-end planning latency, approximately 10\% below the 190 ms fixed block-15 planner and 47\% below the 320 ms fixed full-depth planner. Code will be released.
Open source S6planning latency
320
View original evidence
n rate. On an A100, adaptive routing improves PDMS from 90.62 to 90.79 while averaging 170 ms end-to-end planning latency, approximately 10\% below the 190 ms fixed block-15 planner and 47\% below the 320 ms fixed full-depth planner. Code will be released.
Open source S6
- code: Reported pending Open source S6
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ng to trajectory quality. On NAVSIM, the adaptive single-trajectory planner achieves 90.8 PDMS; a separate fixed-exit variant reaches 92.6 PDMS with 64 proposals. It further obtains 89.9 EPDMS on NAVSIM v2, yielding the best reported results among the compared front-view video world-model planners. Without target-domain fine-tuning, Adaptive-WAM transfers to nuScenes with 0.88 m average L2 error and a 0.08\% collisio
Open source S5
n rate. On an A100, adaptive routing improves PDMS from 90.62 to 90.79 while averaging 170 ms end-to-end planning latency, approximately 10\% below the 190 ms fixed block-15 planner and 47\% below the 320 ms fixed full-depth planner. Code will be released.
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org