Long Chen

Frontier generalist computer-use agents, previously closed-source, are now open for systematic study and community development, with the first release achieving significant performance gains.

Training GUI agents in mocked environments yields substantial real-world performance gains, bridging the gap between simulation and physical device interaction.

Winning the IROS 2026 Indoor Exploration Competition, this solution integrates pre-trained map completion and time-aware homing for single-robot exploration, and a utility-driven, data-sharing approach for multi-robot coordination.

Robots can master delicate tool manipulation directly from unpaired human videos, achieving a 73% performance leap over SOTA by anchoring control in object-centric 3D pose priors rather than raw end-to-end pixels.

Treating all latent tokens identically bottlenecks multimodal test-time compute—disentangling hidden-state updates into visual-grounding and high-entropy reasoning channels beats standard output-space search under matched sample budgets.

Pre-indexing hours of video is a computational dead end—growing an online, query-driven textual memory tree during inference matches offline accuracy without the upfront processing tax.

A training-free framework that reformulates failure recovery as test-time scaling over causal histories, which decomposes recovery into three coupled decisions: when to revise the causal history, where to recover a reliable history prefix, and which history configuration best supports subsequent execution.

Current agent runtimes crater to sub-30% success the moment repairs demand propagating implicit dependencies across interconnected system artifacts rather than patching isolated bugs.

Video models consistently fail at realistic dynamics because they conflate appearance with physical laws, a failure resolved here by synthesizing material-conditioned optical flow fields before rendering pixels.

Current LLMs falter in mobile environments, with performance plummeting under real-world constraints, revealing critical gaps in their planning capabilities.

Qwen-UI-Agent outperforms leading models in mobile and cross-platform tasks, achieving up to 97.5% accuracy on key benchmarks.