signal量子位2026-09-25
After a Decade, AI Expert Signs New Paper
Ren Shaoqing, known for Faster R-CNN and ResNet, returns as corresponding author on a new autonomous driving paper. The proposed MM-Future model generates multiple paired scene-action hypotheses simultaneously, achieving 94.0 PDMS on NAVSIM-v1 and 91.5 EPDMS on NAVSIM-v2. It also improves training convergence, reaching a 0.80 validation PDM score in 3.8k steps versus 17.5k steps for single-mode.
- for who
- Researchers and engineers working on world models, planning, and end-to-end autonomous driving.
- why now
- AI大牛任少卿时隔十年再署名，蔚来自动驾驶新论文发布。
- what changes
- World models now extend from predicting future scenes to actively comparing outcomes of different actions, shifting multi-modal planning from trajectory diversity to joint action-future diversity.
- to do
- Read the MM-Future paper and consider implementing its multi-mode joint modeling approach in planning pipelines.
key points
- MM-Future jointly models multiple scene-action hypotheses, enabling 'walk one step, think ten steps'.
- Achieves 94.0 PDMS on NAVSIM-v1 and 91.5 EPDMS on NAVSIM-v2, outperforming prior WAM and E2E baselines.
- Sampling 64 candidates adds ~233ms latency on H800; multi-mode training converges 4.6x faster than single-mode.
#autonomous driving#world model#end-to-end#multi-modal planning#NIO
score
score 9 out of 10. 0-10: how dense the facts are, multiplied by how much you can do with them after reading. 8+ means the topic's evidence bar is met: benchmarks and availability for a new model, amount and investors for a funding round, revenue figures for a solo-money story. Below 5 an item does not enter the digest. A press release scores 3 or less, a reprint loses 2, anything older than 14 days loses 1, a headline that misleads loses 3.
read the source