Junyang Lin

Qwen-VLA introduces a unified embodied foundation model that extends vision-language understanding to continuous action generation, enabling robust, multi-task generalization across diverse robotic tasks and embodiments.

Qwen-VLA introduces a unified embodied foundation model that extends vision-language understanding to continuous action generation, enabling robust, multi-task generalization across diverse robotic tasks and embodiments.

The paper presents Bebop, a method to improve the rollout stage in reinforcement learning by optimizing Multi-Token Prediction (MTP) acceptance rates.

This paper proposes a unified autoregressive framework (UniAR) for integrating visual understanding and generation within a single system.

Qwen-RobotNav is a scalable navigation model that can be reconfigured for different tasks and environments.

The paper proposes OmniAgent, a native omni-modal agent that formulates video understanding as a POMDP-based iterative Observation-Thought-Action cycle, enabling active perception and improving performance with increasing reasoning turns.