CCTest · Blog
CoPE Trains Language Agents to Act and Forecast Together
AI Agents
cctest.ai
AI Agents

CoPE Trains Language Agents to Act and Forecast Together

CoPE adds multi-step action forecasting to agentic reinforcement learning, allowing successful trajectories to provide supervision before the terminal reward. Evaluations across five benchmarks show stronger aggregate results than GRPO and related predictive-auxiliary baselines, with sample efficiency reaching up to roughly 2× that of some baselines.

Read more
CCTest · Blog
BlenderWorld Tests Whether Visual Agents Can Really Create in 3D
Evaluation & Benchmarks
cctest.ai

BlenderWorld Tests Whether Visual Agents Can Really Create in 3D

BlenderWorld introduces a benchmark for evaluating visual agents on realistic, open-ended 3D creation tasks. Its results show that frontier models can understand broad instructions, but still struggle with spatial coherence, structural reconstruction, and complete editable Blender deliverables.

Read more
CCTest · Blog
AffordanceWAM Turns Human Videos into a Shared Interface for Robot Manipulation
Robotics & Physical AI
cctest.ai

AffordanceWAM Turns Human Videos into a Shared Interface for Robot Manipulation

AffordanceWAM introduces an affordance-aware joint world-action model that learns objects, interaction regions, and scene evolution from human videos before generating robot actions from limited robot trajectories. It avoids human action labels and motion retargeting.

Read more