As you know, we are organising a #Neurips2023 competition on robotics, where we are putting all those #MachineLearning claims to the test! The real-test people! I wrote a blog on the competition and the results so far we have on the leaderboard! Check it out:

#ICCV2023 Traj-MAE: Masked Autoencoders for Trajectory Prediction. Joined work with my co-supervised PhD students at CUHK. jiazewang.com/projects/trajm…

#NeurIPS2023 Happy to share our new NeurIPS 2023 paper: ChessGPT: Bridging Policy Learning and Language Modeling. We present how we conduct policy learning and language modeling in GPT model simultaneously in Chess! Paper: openreview.net/forum?id=pvdm4… Code: github.com/waterhorse1/Ch…

Strongly welcome to this event and look forward to exchanging LLM, MLLM, AI Agent with you in Vienna!

[1/3] Happy to share our new GUI Agent paper: SPA-Bench: A Comprehensive Benchmark for Smartphone Agent Evaluation. Key contributions: (1) A diverse set of tasks; (2) A plug-and-play Agent framework; (3) A novel evaluation pipeline! Project page: ai-agents-2030.github.io/SPA-Bench/

[2/3] Happy to share our new GUI Agent paper: Lightweight Neural App Control. We propose LiMAC, an architecture that balances efficiency and natural language understanding by combining a lightweight transformer with a fine-tuned VLM. Project page: arxiv.org/abs/2410.17883

We are #hiring for AI Agent research scientist, research engineer, research intern at Huawei Noah’s Ark Lab. Message me if you're interested in joining our team. We are attending Conference on NeurIPS 2024 if you would like to meet! #NeurIPSConference

#ICLR2025 Kicking off with GUI Agents' real-world needs, we've explored action model design, reinforcement fine-tuning, and benchmarking. If you're into GUI Agents, let's collaborate! Full-time and internship positions are open. #GUIAgents #AIAgent

Excited to share that our paper "Lightweight Neural App Control (LiMAC)" is accepted to #ICLR2025! 🎉 LiMAC enables app control on mobile devices using small architectures (<2B params), offering a lightweight alternative to large frameworks like OpenAI's Operator. 🧵👇 (1/3)

I'll attend ICLR 2025 in Singapore! Three papers (2 spotlight) at the main conference. If you want to chat about RL/Agent/LLM, pls DM me! #ICLR2025

🚀 ViMo – The First Visual World Model for App Agents 📍 Project page: ai-agents-2030.github.io/ViMo/ 🎯 ViMo fills a critical gap by bringing vision-language understanding into mobile interaction, enabling agents to simulate, reason, and plan in mobile environments.

🚀 VSC-RL, New VLM agents 🤖 for resolving mobile device 📱 and web 🖥️ control tasks. 🔗Website: ai-agents-2030.github.io/VSC-RL/ VSC-RL autonomously decomposes complex goals into feasible subgoals, boosting learning efficiency and performance.

Deep Research Agents: A Systematic Examination And Roadmap github.com/ai-agents-2030…

Delighted to announce that three of our papers have been accepted at #NeurIPS2025 ! This was a true team effort, and I want to thank all of our collaborators for their invaluable contributions. Onward and upward in the world of AI Agents and Agentic RL!

Happy to share that we've had 3 papers accepted at #AAAI2026! A huge thank you to all our collaborators—this success wouldn't be possible without your incredible contributions!

Darwin Mobile Agent: the 1st FULLY OPEN-SOURCE, end-to-end pipeline for large-scale online RL and inference of mobile GUI agents. ai-agents-2030.github.io/Darwin-Mobile-… #ReinforcementLearning #GUIAgents #DarwinAgent

#ICLR+3 🎉 Thrilled to announce that ViMo has been accepted! As the first multimodal world model in the GUI Agent domain, it further validates the emerging paradigm for future personal AI agents: GUI Agent + GenUI.