CMU paper: one computer-use agent ranks top-3 across four embodied ECCV competitions
wellecks · x · 2026-11-07
A position paper by Pranjal Aggarwal and Sean Welleck (CMU) argues computer-use agents are embodied generalists. A single CUA entered four embodied AI competitions at ECCV workshops, ranking 1st, 2nd, 3rd and 6th — with no agent tuning, only simple GUI software rendering each benchmark (e.g., navigating a 3D house via keyboard in a web interface). The claim: embodied tasks can be formulated as digital software interaction, so skills CUAs learn on software — visual grounding, precise clicking, exploration — already transfer to embodied tasks. The authors advocate embodied AI can converge on a single general model using the same training strategy that made computer use general.
More from Embodied
- Hugging Face hosts ~100K open robotics datasets, up 48% in six months — micoolcho · 2026-11-07
- Walden Robotics teases "exciting AI news" at CoRL 2026 keynote "Learning in the factories" — adnothing · 2026-11-07
- YC-backed Trepo debuts kitchen hardware that lets AI agents shop for groceries in the background — ycombinator · 2026-11-07
- FlexiTac-Gen launches: AI-agent-assisted tool to design custom tactile sensors, accepted to CoRL 2026 — Haoyu_Xiong_ · 2026-11-07
- PolicyIK: NYU trains RL for inverse kinematics on a real arm, hitting 0.5mm median error — Lianhuiq · 2026-11-07
- Unitree factory tour sparks debate: robot labor is 8% of costs in Hangzhou, ~40% at US wages — teortaxesTex · 2026-11-07