Li et al.
Lin Li is a researcher at Robbyant, Ant Group's embodied artificial intelligence subsidiary, where they focus on foundation models for robot learning and control. Li is the lead author of LingBot-VA, formally titled "Causal World Modeling for Robot Control" and accepted at Robotics: Science and Systems (RSS) 2026, which introduces a causal video-action world model that jointly learns frame prediction and action execution via an autoregressive diffusion framework. The work is best known for demonstrating strong bimanual manipulation capabilities through large-scale cross-embodiment data pre-training and an asynchronous inference pipeline that decouples action prediction from motor execution.
“LingBot-VA hits 74.2% success on the bimanual RoboTwin 2.0-Plus benchmark.”
Source→“LingBot-VA uses 16,000+ hours of cross-embodiment robot data. The 'data efficiency' framing is accurate only for the fine-tuning stage.”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.