Reproduce “Thinker-WM: A Unified World Model with Shared Multimodal Representations for Embodied Intelligence” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Thinker Team Ubtech, Peng Gao, Jichao Jiao, Chen Liao, Daqin Luo, Yiming Ma, Baiyu Pan, Hailin Shi and 3 more
Thinker-WM jointly models robot video, metric depth, and action using modality-specific encoders and decoders around a shared dynamics model. Depth is one-way during pretraining, with bidirectional interaction enabled in post-training. On EWMBench, it scores 4.66 versus 4.60 for Qwen-RobotWorld; a simulated RoboTwin task also favors bidirectional interaction, but real-robot closed-loop performance remains unestablished.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.