Jia Xiaosong
Assistant Professor at the Institute of Trustworthy Embodied AI, Fudan University, researching autonomous driving and embodied intelligence. He has proposed models including TCP, ThinkTwice and DriveAdapter, and focuses on end-to-end autonomous driving and imitation learning/reinforcement learning training paradigms.
Timeline (last 90 days)
Jia Xiaosong is an assistant professor at the Institute of Trustworthy Embodied Intelligence, Fudan University.
Jia Xiaosong used ThinkTwice to run a large number of closed-loop experiments in the CARLA simulator, collecting nearly unlimited data and investing 100TB of storage and substantial compute to train larger models, but driving capability did not jump from "drives okay" to "drives perfectly" as data and model scale grew.
Jia Xiaosong believes imitation learning can take a model from 0 points to 90 points and reinforcement learning can tackle the remaining 10 points, and the two are not substitutes for each other; reinforcement learning starting completely from scratch converges extremely slowly.
Jia Xiaosong observed that models such as VLA and WAM that introduce stronger vision-language capabilities are not necessarily better than well-optimized end-to-end models on some autonomous driving benchmarks; what truly limits the model is whether it can make the right decision, not whether it recognizes a certain object.
Jia Xiaosong believes that if training remains dominated by imitation learning, changing the architecture cannot bypass the inherent problems of imitation learning; valuable directions include filtering valuable data, going beyond brute-force imitation learning, designing more effective reward signals, and enabling models to reinforcement-learn tens of thousands of skills simultaneously.
Jia Xiaosong believes embodied intelligence is much harder than autonomous driving: contact, friction and deformable objects are difficult to model precisely, the hallucination and speed problems of world models also exist, and the methodology of autonomous driving can be borrowed but one cannot assume it has solved the problems of embodied intelligence.
All Facts (6)
Jia Xiaosong is an assistant professor at the Institute of Trustworthy Embodied Intelligence, Fudan University.
50%UnverifiedJia Xiaosong used ThinkTwice to run a large number of closed-loop experiments in the CARLA simulator, collecting nearly unlimited data and investing 100TB of storage and substantial compute to train larger models, but driving capability did not jump from "drives okay" to "drives perfectly" as data and model scale grew.
50%UnverifiedJia Xiaosong believes imitation learning can take a model from 0 points to 90 points and reinforcement learning can tackle the remaining 10 points, and the two are not substitutes for each other; reinforcement learning starting completely from scratch converges extremely slowly.
50%UnverifiedJia Xiaosong observed that models such as VLA and WAM that introduce stronger vision-language capabilities are not necessarily better than well-optimized end-to-end models on some autonomous driving benchmarks; what truly limits the model is whether it can make the right decision, not whether it recognizes a certain object.
50%UnverifiedJia Xiaosong believes that if training remains dominated by imitation learning, changing the architecture cannot bypass the inherent problems of imitation learning; valuable directions include filtering valuable data, going beyond brute-force imitation learning, designing more effective reward signals, and enabling models to reinforcement-learn tens of thousands of skills simultaneously.
50%UnverifiedJia Xiaosong believes embodied intelligence is much harder than autonomous driving: contact, friction and deformable objects are difficult to model precisely, the hallucination and speed problems of world models also exist, and the methodology of autonomous driving can be borrowed but one cannot assume it has solved the problems of embodied intelligence.
50%