Tencent Releases Embodied Multimodal Foundation Model RxBrain
jacek2023 · reddit · 2026-07-15
This HF page introduces Tencent's Hy-Embodied-RxBrain-1.0, a unified multimodal foundation model designed for embodied cognition.
What It Does
- Embodied Understanding & Reasoning: Performs Q&A and chain-of-thought reasoning on images and multi-frame videos.
- World State Prediction: Imagines what visual frames will appear next in the physical world based on actions.
- Joint Sub-goal Planning: Breaks tasks into steps, simultaneously outputting the "next action" and a "target image" for each step.
Architecture Highlights
- Employs interleaved generation: alternately generating text reasoning and imagined frames within the same autoregressive sequence.
- Uses a single <Image> token to decide when to enter the "imagination" phase.
- The backbone is a unified Mixture-of-Transformers (MoT) with approximately 6.2B parameters.
- The image imagination component decodes into the latent space of a frozen FLUX VAE via a flow-matching image head.
Design Philosophy Emphasized
Rather than splitting understanding, generation, and planning into separate towers, it attempts to couple "what to say" and "what the world should look like" within the same sequence.
More from Embodied
- Johns Hopkins Launches Full-Stack Hands-on Robot Learning Class with SO-101 Arm Kits — _krishna_murthy · 2026-09-11
- SyncWorld: In-Context Robot World Model Simulates Unseen Views and Embodiments Zero-Shot — ChongZzZhang · 2026-09-11
- A 3D Pose Dataset for Dogs Released — ducha_aiki · 2026-09-11
- Swaayatt demos autonomous driving at 52 km/h on mountain roads, self-recovers after skid — sanjeevs_iitr · 2026-09-11
- AUAR's MicroFactory brings a deployable robotic wood-panel factory to the construction site — lukas_m_ziegler · 2026-09-11
- Musk: Cybercab certified at 165 Wh/mi, the most efficient production EV ever — elonmusk · 2026-09-11