Explore cutting-edge advancements in robot learning and manipulation this week. We dive into new world models like DVG-WM and AdaJEPA that boost efficiency and adaptability for complex tasks. Discover how robots are learning from freeform human preferences and even translating human video demonstrations into humanoid control with Human-as-Humanoid. Plus, uncover how metacognitive feedback in reinforcement learning can make LLMs express uncertainty more faithfully.

Explore the cutting edge of generative AI as we dive into groundbreaking research transforming robotics and language models. Discover how DanceOPD and World Action Models are pushing the boundaries of continual imitation learning, alongside new approaches

Discover how researchers are making Multimodal LLMs fairer judges by mitigating perceptual bias and enhancing their continuous learning capabilities for new instructions. Explore groundbreaking methods for scalable robot data synthesis with compositional world models and a new predictive visual code for efficient video processing by MLLMs. Plus, delve into an innovative interactive environment simulating multi-stage clinical scenarios using electronic health records for robust AI agent training.