Research
We are building the foundational capabilities for general-purpose embodied AI — systems that can perceive, reason, and act in the physical world.
Latest Research
LWD
Learning While Deploying turns real-world robot deployment into a continual reinforcement learning loop, where a shared generalist VLA policy improves from the experience collected by a robot fleet.
A fleet-scale reinforcement learning framework for continual post-training of generalist robot policies.
Apr 30, 2026Learn More
SOP
Scalable Online Post-Training for VLA Models
A framework that enables robots to continuously improve through real-world online post-training at scale.
Jan 6, 2026Learn More
Act2Goal
From World Model to General Goal-Conditioned Policy
A unified auto-regressive framework that jointly pre-trains vision-language reasoning and robotic action generation.
Jan 1, 2026Learn More
UniFact
Unified Embodied VLM Reasoning with Robotic Action
Bridges world model prediction and robot action, enabling robots to achieve arbitrary goals without task-specific fine-tuning.
Jan 1, 2026Learn More
