I'm blessed to share our latest work at the NTU AI-CoRE Top Conference Meetup. Praise the Lord!
📄 Paper: https://arxiv.org/abs/2607.01225
We study imitation learning from suboptimal demonstrations using language instead of traditional scalar supervision.
No Digg Deeper questions have been answered for this story yet.
Excited to share our new paper on learning from suboptimal demonstrations with language feedback, led by @ch_cyr_25 Chih-Han Yang!
If you’re interested in how we generate language feedback and train agents to learn from it—and you’re attending ICML—chat with Chih-Han!
I'm blessed to share our latest work at the NTU AI-CoRE Top Conference Meetup. Praise the Lord!
📄 Paper: https://arxiv.org/abs/2607.01225
We study imitation learning from suboptimal demonstrations using language instead of traditional scalar supervision.
I’ll also be at ICML 2026—happy to chat if you’re around!
Joint work with @jaywu109 , Yun-Ping Huang , @PCHsiehtw @Kenneth_Marino @shaohua0116
#MachineLearning #RobotLearning #ImitationLearning
We propose a language generator and a Language-Critique (LC) loss for end-to-end policy learning. Across 8 continuous-control tasks, our method achieves competitive or better performance than existing baselines.
Previous methods rely on signals like confidence scores or Q-values, but these cannot explain why an action fails or how to improve it. Language can provide richer supervision, including task progress, failure modes, and corrective guidance.