# Robotics — AI updates Embodied AI — robots, simulation, and the systems that train them for the physical world. - [Kyutai releases REKI, a 2.4B robot model with efficient memory](https://www.parallelquant.com/posts/kyutai-releases-reki-a-2-4b-robot-model-with-efficient-memory-2131b0) (2026-10-17, MarkTechPost): Kyutai, with Inria and École des Ponts, released REKI, a 2.4B-parameter vision-language-action (VLA) model for robot manipulation. It is pretrained only on public data and keeps a cheap memory of its recent past. On three out-of-distribution benchmarks, only models about twice its size beat it, and all weights are public. - [Small AI models let drones autonomously target and attack in combat](https://www.parallelquant.com/posts/small-ai-models-let-drones-autonomously-target-and-attack-in-combat-3b4bc8) (2026-09-17, Ars Technica): A company called Scaleout is deploying decentralized AI-driven learning to military bases and drones, using small on-device models that let drones identify and attack battlefield targets without a human in the loop at the moment of engagement. - [Agility's new humanoid robot can squat to avoid hitting people](https://www.parallelquant.com/posts/agility-s-new-humanoid-robot-can-squat-to-avoid-hitting-people-1bb6fa) (2026-09-15, Ars Technica): Agility Robotics unveiled a humanoid robot designed to stop or crouch when it detects a nearby human coworker, rather than requiring a safety cage. The goal is to let the robot work directly alongside people on a factory or warehouse floor. - [Reward AI trains robot manipulation policy from human demos only](https://www.parallelquant.com/posts/reward-ai-trains-robot-manipulation-policy-from-human-demos-only-2ede71) (2026-09-14, MarkTechPost): Reward AI released OM-1, a general-purpose manipulation policy trained solely on human demonstrations captured via a 7-degree-of-freedom wearable glove, with no teleoperation or on-robot data used. The policy runs on industrial arms and humanoids at human speed and can learn a new task from under 30 minutes of data. No weights, code, or API have been released publicly yet. - [Maven Robotics emerges from stealth with $100M and live deployments](https://www.parallelquant.com/posts/maven-robotics-emerges-from-stealth-with-100m-and-live-deployments-dd5079) (2026-09-10, TechCrunch Startups): Maven Robotics came out of stealth with a $100 million Series A, saying it already has active robot deployments and is positioning itself to compete for robotics deployment contracts with established players. - [GPT-6 Astra shows early gains in robot spatial reasoning](https://www.parallelquant.com/posts/gpt-6-astra-shows-early-gains-in-robot-spatial-reasoning-7a7a47) (2026-09-12, The Decoder): On a new robotics benchmark called StationeryBench, GPT-6 Astra completed 7 out of 100 dual-arm robot tasks, while competitor MolmoAct2 completed zero. A researcher described the jump as a "step change" in spatial reasoning. - [Skild AI's robot model learns new tasks from a single video](https://www.parallelquant.com/posts/skild-ai-s-robot-model-learns-new-tasks-from-a-single-video-f5ae01) (2026-09-10, NVIDIA): Skild AI's S1 robot foundation model, built with Nvidia's Physical AI tools, can learn previously unseen, long-horizon tasks from a single video demonstration rather than extensive reprogramming. It targets manufacturing and warehouse settings where layouts and tasks change frequently. - [Maven Robotics emerges from stealth with $100M and live deployments](https://www.parallelquant.com/posts/maven-robotics-emerges-from-stealth-with-100m-and-live-deployments-40a475) (2026-09-10, TechCrunch): Maven Robotics came out of stealth with a $100 million Series A funding round. The startup already has active robot deployments in place. - [Browser-based tool collects 50,000+ robot manipulation demonstrations](https://www.parallelquant.com/posts/browser-based-tool-collects-50-000-robot-manipulation-demonstrations-923bc2) (2026-09-07, MarkTechPost): Axis Robotics released AXIS, a browser-based data collection system that moves robot demonstration gathering out of physical labs onto GPU-backed backends. The dataset includes 207 manipulation tasks and 50,129 verified trajectories on a Franka arm. Pretraining on it lifted the pi-0.5 robot policy's LIBERO-Plus score from 83.9 to 88.8, versus 57.5 for a volume-matched control dataset. - [Humanoid robots break Usain Bolt's 100m sprint record](https://www.parallelquant.com/posts/humanoid-robots-break-usain-bolt-s-100m-sprint-record-a2484d) (2026-08-23, Tom's Hardware): At the 2026 World Humanoid Robot Games in Beijing, the Tiangong Ultra robot ran 100 meters in 9.39 seconds, and the Honor Lightning followed at 9.47 seconds — both beating Usain Bolt's 9.58-second world record. The Tiangong Ultra reportedly reached 23.8 mph before crashing into padded barriers at the finish. --- Published by Parallel Quant — https://www.parallelquant.com