CCTest · Blog
RynnVLA-Latent Scales Robot Policies with 3D-Grounded Latent Actions
Robotics & Physical AI
cctest.ai

RynnVLA-Latent Scales Robot Policies with 3D-Grounded Latent Actions

RynnVLA-Latent introduces a way to pretrain vision-language-action models from heterogeneous videos without requiring native robot action labels. Its 3D-grounded latent actions provide a shared interface between human demonstrations, robot trajectories, simulations, and different embodiments.

Read more
CCTest · Blog
RoboHarness: A Tool Interface for LLM-Driven Robot Control
Robotics & Physical AI
cctest.ai

RoboHarness: A Tool Interface for LLM-Driven Robot Control

RoboHarness explores an alternative to training ever-larger vision-language-action models: give an LLM agent a visual-geometric control interface and let it invoke embodied tasks directly. The paper reports that RoboHarness with Qwen3.8-Flash-Next outperformed state-of-the-art VLA models on BEHAVIOR Challenge 2025 tasks.

Read more
CCTest · Blog
iPrefBench Tests Whether Mobile Agents Know When to Infer or Ask
Evaluation & Benchmarks
cctest.ai

iPrefBench Tests Whether Mobile Agents Know When to Infer or Ask

iPrefBench evaluates personal mobile-use agents in an executable iOS environment, testing whether they can infer user preferences from activity history and ask for information that cannot be inferred. The results expose a wide gap between competent UI execution and genuinely personalized task completion.

Read more