Jia Xiaosong, an assistant professor at Fudan University, has navigated significant shifts in autonomous driving technology over the past five years. Initially focused on trajectory prediction, he transitioned to end-to-end driving models, then to reinforcement learning, and most recently to world models. His research, including projects like Think2Drive and Bench2Drive-R, highlights the limitations of traditional modular systems and imitation learning, advocating for approaches that directly optimize final objectives and simulate interactions in latent spaces to reduce training costs. Jia believes that while new model architectures are emerging, the true bottlenecks in autonomous driving and embodied AI lie in data efficiency and the fundamental challenges of learning from interaction, rather than just model structure. AI
IMPACT Highlights that true AI progress in embodied systems may hinge on data efficiency and interaction learning, not just model architecture.
RANK_REASON Interview with a researcher discussing trends and challenges in autonomous driving and embodied AI.
- Bench2Drive-R
- CARLA Leaderboard 2.0
- European Conference on Computer Vision
- Fudan University
- Jia Xiaosong
- Li Hongyang
- Masayoshi Tomizuka
- Nvidia
- Shanghai Jiao Tong University
- Think2Drive
- University of California, Berkeley
- Yan Junchi
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
