MuJoCo (Product): news timeline & CVEs

Google wants Gemini 4 Argon to be the only AI model robots need

Google pitches Gemini 4 Argon as one robotics model for control, simulation, and training-data grading.

Google is pitching Gemini 4 Argon as a single model for robot control, simulation, and training-data review, including demos with a Boston Dynamics Spot robot in MuJoCo. From one photo it generated 77 tasks and more than 4,500 episodes, raising real-robot success from 56% to 75% and cutting execution from 35 to 24 seconds. Google says Argon scored 87% on the 400-question ERQA benchmark, ahead of Claude Opus 5.5 at 82.3%, GPT-6 Astra at 81%, and Claude Fable 5.1 at 74%. On Artificial Analysis’s AA-Robotics test it scored 55, behind GPT-6 Astra at 69 and GPT-6.1 Sol at 65.

The Decoderupdated Oct 16, 15:03 UTC · 10h agofirst Oct 15, 16:35 UTC · 1d agoModel release 5 sources3· 3 reads

Gemini 4 Argon for Robotics

Google launched Gemini 4 Argon, a robotics model scoring 87% on ERQA.

Google introduced Gemini 4 Argon, a general-purpose physical-reasoning model aimed at robotics. On the Embodied Reasoning (ERQA) benchmark it scores 87%, which Google says exceeds other frontier models. Demonstrations include Boston Dynamics Spot mobile manipulation via MuJoCo practice, zero-shot robot-arm control, closed-loop steering of a video world model, Real2Sim reconstruction, and iterative self-improvement.

Hacker News · AIupdated Oct 16, 15:03 UTC · 10h agofirst Oct 13, 16:55 UTC · 3d agoModel release 5 sources HN 232↑ · 57 comments

Gemini 4 Argon for Robotics

Gemini 4 Argon scores 87% on ERQA and demonstrates new robotics control workflows.

Google DeepMind says Gemini 4 Argon improves physical understanding, code generation, and visual-semantic reasoning for robotics. Using pure reasoning, Argon scores 87% on the Embodied Reasoning (ERQA) benchmark, ahead of other frontier models in the post. Demonstrations include Boston Dynamics Spot mobile manipulation practiced in MuJoCo, zero-shot robotic-arm control, and closed-loop steering of the Gemini Omni world model. The post also covers real-to-sim reconstruction, dexterity labeling, and iterative improvement in simulation and on hardware.

Google DeepMindupdated Oct 16, 15:03 UTC · 10h agofirst Oct 13, 16:01 UTC · 3d agoAI research 5 sources

Can Safeworld convince people that gen AI robots won’t hurt them?

Safeworld exits stealth with over $12 million to simulate and underwrite generative-AI robot safety.

Safeworld, founded by Carnegie Mellon Safe AI lab director Ding Zhao with Kyle Wong and Simo Rachidi, emerged from stealth to evaluate whether generative-AI robot controllers are safe to deploy. The company raised a seed round of more than $12 million led by Shine Capital and a16z Speedrun, with Box Group, the Carnegie Mellon University Endowment, Innovation Endeavors, and SV Angel. It rebuilds site-specific scenes in simulators such as Genesis or MuJoCo and runs thousands of scenarios with realistic human models, including falls, occluded corners, and varied postures. Construction-robot company Gritt Robotics is an early partner, and Safeworld has not settled whether it will sell a platform or services.

TechCrunch · AIOct 5, 12:00 UTC · 11d agoAI industry2

FERPO: Forward Entropy-Regularized Policy Optimization

FERPO updates continuous-control policies from critic values without differentiating the critic by actions.

FERPO is an on-policy maximum-entropy reinforcement learning algorithm that derives an entropy- and KL-regularized target action distribution and fits the actor by minimizing a forward-KL objective. The objective is estimated with self-normalized importance sampling from the rollout policy, avoiding differentiation of the critic with respect to actions. Experiments on MuJoCo Playground and ManiSkill show competitive performance and sample-efficiency gains, and actor updates are faster than Relative Entropy Pathwise Policy Optimization.

arXiv cs.AI / cs.LG / cs.CLOct 1, 17:59 UTC · 15d agoAI research2· 1 read

MuJoCo

Timeline5 articles

Appears with

Google wants Gemini 4 Argon to be the only AI model robots need

Gemini 4 Argon for Robotics

Gemini 4 Argon for Robotics

Can Safeworld convince people that gen AI robots won’t hurt them?

FERPO: Forward Entropy-Regularized Policy Optimization