Frontier Language Models Accelerate Robotics Research Across Simulation and Training
- Published
- Source
- HumanoidsDaily
Ros Shinkle, founder and CEO of Tuesday Labs and a former staff roboticist at Boston Dynamics, said that before October 20, training a whole-body controller took one week on 128 GPUs, while it now takes three minutes on one GPU. Andrew Kang, CEO of RoboStrategy, attributed the broader acceleration to algorithmic optimizations, robot-data efforts, new scaling laws, advances in world models and simulation, open-source tools, and language models automating more of the research process.
A sequence of open-source projects illustrates the speed gains. UC Berkeley’s SuperMimic paper, posted to arXiv on October 20, describes training humanoid motion-tracking policies through a learned, differentiable world model; its authors report that three hours on one GPU outperforms a tracker trained for seven days on 128 GPUs. After an unofficial ETH Zürich reimplementation appeared the next day, projects including InstantTrack, BlinkMimic, and XMimic reduced Unitree G1 training time on the shared 40-clip LAFAN1 benchmark from 1.31 hours on one GPU to about 2.5 minutes on one H100, or 0.86 minutes on four GPUs, over 17 days.
Language models are also being used to create simulation environments and tasks. Marius Memmel reported that Astra generated 25 environments, 25 multi-step tasks, and five robot embodiments in a 23-hour run. Vector Wang said Claude Opus 5.5 recreated a laundry task in browser-based MuJoCo and trained a reinforcement-learning whole-body controller in three hours. The extent of AI assistance in the later project repositories was not disclosed.
Original report
How Frontier Language Models Are Speeding Up Robotics Research
HumanoidsDaily · [email protected] (Humanoids Daily Staff)This briefing is an independently written summary based on publicly available reporting and is provided for industry information and news discovery. The original report and source publication are credited and linked where applicable. RobotToday does not claim ownership of third-party source material.
Rights concerns? Contact [email protected] with the relevant URL and details. We will review the matter and take appropriate action where warranted.
More related news
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
Related Suppliers
Tuesday Lab
New York-based robotics startup building delightful home tidying robots with supervised autonomy, founded by former Boston Dynamics engineers, YC W24.
Hybrid Robotics Lab, UC Berkeley
UC Berkeley lab led by Prof. Koushil Sreenath researching nonlinear control of legged and aerial robots; creator of the open-source Berkeley Humanoid Lite.
UC Berkeley Robotics and Intelligent Machines / BAIR Lab
UC Berkeley's AI and robotics research lab with 50+ faculty; known for robot learning, manipulation, and reinforcement learning research by Abbeel.
Seer Robotics 上海仙工智能
HKEX-listed Shanghai maker of SRC mobile-robot controllers and software, selling the navigation brain that integrators build their own AMRs around.
Shenzhen Yuhesen Technology Co., Ltd 深圳煜禾森科技有限公司
Shenzhen-based maker of modular AGV/AMR chassis platforms and SLAM navigation modules for logistics, agriculture and inspection robots.
Delta Intelligence 德塔智能
Beijing embodied-AI startup founded 2026 building the Delta-0 humanoid foundation model and D1 data-capture rigs; about RMB 1.2bn raised in six rounds.
Maxinsights
San Francisco vendor, founded 2024, of annotated egocentric video, hand/body mocap and tactile data for training robot VLA and world action models.
Learning Robots
Learning Robots is a Paris-Saclay CNRS spin-off making AlphAI, an educational robot and software for teaching artificial intelligence.
BridgeDP Robotics 桥介数物
Shenzhen developer of universal robot 'cerebellum' motion-control algorithms for humanoid, quadruped and wheeled-legged robots using RL and motion capture.
Boardwalk Robotics
Pensacola-based humanoid robotics company developing Alex, a 19-DOF legless upper-torso humanoid for manufacturing and logistics.
Cyborg Robotics 赛博格机器人
Shenzhen humanoid robot maker focused on embodied intelligence for hazardous industry; produces full-size biomimetic humanoids and dexterous hands.
The Robot Learning Company
YC X25 startup building open-source AI-native robot arms; TRLC-DK1 bimanual dev kit enables imitation learning and teleoperation for around USD 1,200.
Robotics needs a service framework.
RSF defines a common language for robot service capability, lifecycle operations, certification pathways, and service-provider networks.
Join the RobotToday community on LinkedIn
Daily robotics news, in-depth analysis, conference highlights, and discussions with professionals worldwide.