Predicts future egocentric video from actions for human and robot embodiments. It represents movement with camera poses and manipulation with projected skeletons, trained on a unified collection of human and robot experience.
AI Agent Papers2026
Evaluates whether agents can search, browse, and combine evidence across the live Web to produce unique factual answers. Its 200 tasks expose persistent weaknesses in long-horizon search and tool coordination.
AI Agent Papers2026
Helps general coding agents turn 3D perception readings into spatial conclusions using persistent scene memory and plain-language interpretation. It improves open- and closed-model results on MMSI-Bench and ReVSI.
AI Dataset2026
Evaluates language models that generate 3D objects from text by comparing their outputs in human preference tests. Scores cover creativity, coherence, prompt alignment, and visual preference.
AI Agent Papers2026
Combines long-horizon task planning with native keyboard-and-mouse control to complete multi-stage Minecraft tasks from visual feedback alone. Structured memory and policy-aware completion checks help the agent verify progress, recover, or replan.
AI Agent Papers2026
Evaluates whether shopping simulators reproduce real-user behavior across long interactive journeys, then checks whether simulated policy effects align with a production A/B test.
AI Agent Papers2026
Zimu Wang, Xiaobo Wang +12University of California, Berkeley, University of Science and Technology of China +5
Builds and compares agent-evolution strategies on shared infrastructure, keeping mutable agent resources separate from protected evaluation. The best observed overall pass@1 gain was 13.5 percentage points, but results varied by setting and did not always transfer to held-out tasks.
AI Agent Papers2026
Evaluates AI agent harness reliability and failure attribution using inspectable execution traces and evidence from multiple sources. Its 95 tasks cover seven dimensions and score task performance separately from harness reliability.
Simulates closed-loop interactions between people and wearable AI assistants to test whether multimodal models can guide real-world tasks. Its 120 multi-turn benchmark tasks are scored against verifiable simulator states.
Trains a spatial vision-language model to transfer across categorical, metric, and temporal reasoning tasks by replaying heterogeneous supervision chronologically. It also evaluates adaptive frame selection as a separate inference-time improvement.
AI Video Papers2026
Infers implicit user intent from continuous video and adapts both response timing and content as goals change. Its benchmark covers explicit, implicit, and dynamic intents, while hierarchical memory tracks state across sustained interactions.
AI Agent Papers2026
Coordinates world creation, trajectory exploration, and Harness evolution to improve multimodal agents. It organizes tasks by shared capabilities rather than surface categories, reporting mean gains of 34 points on training benchmarks and 26 on unseen benchmarks.