Tag
BeingBeyond is gathering precise training data for humanoid robots by attaching robotic hands alongside human hands, likely for teleoperation-based learning.
Sergey Levine highlights a new paper investigating why action chunking is so effective in modern large-scale imitation learning for robotics, breaking down the underlying reasons.
Google DeepMind shares an interview with Apollo 2, a humanoid robot, about its experience running on the Gemini Robotics 2 model.
A tweet outlines Elon Musk's vision of building a massive Terafab chip manufacturing facility to power AI, robotics, and automation, turning AI into physical infrastructure that could reshape the 2030s.
TRACE introduces a novel approach to active 3D reconstruction by optimizing full sensor trajectories for ergodic coverage of scene information, outperforming next-best-view baselines with a 1.5 dB PSNR improvement.
A researcher muses on the unstable equilibrium where language uses autoregressive models while other modalities use diffusion, and speculates that a unified multimodal architecture depends on the order each modality reveals information. He seals a bet on this idea.
Persona.AI demos its Gen 1 humanoid robot performing welding tasks via teleoperation, targeting dangerous jobs.
Macrodata Labs releases a research blog on scaling robotics with egocentric video data by recovering 3D hand motion signals using only open-source models.
China's humanoid robot leader Unitree is raising $904M in a mainland IPO at a $9B valuation, having shipped over 5,500 humanoids in 2025 and partnering with DeepSeek on model development.
Chinese humanoid robots perform Webster flips and side aerials on a martial-arts set, using real-time balance algorithms to keep the 55 kg machine stable through the routine.
NVIDIA highlights the importance of open world models for physical AI, introducing the NVIDIA Cosmos 3 open model family and Omniverse libraries for simulation and model specialization.
Jensen Huang announced that NVIDIA has officially open-sourced its autonomous driving reasoning model Alpamayo 2 Super. The model can understand complex scenes and think before acting, and is suitable for Robotaxi, trucks, delivery vehicles, etc. It is open for commercial use under the OpenMDW-1.1 license.
NeuroPB is a framework that scales neural decoding by pretraining a motor encoder on large-scale behavioral data (including robotic trajectories) and aligning neural activity to that representation space, improving trajectory decoding and generalization with limited neural data.
This paper presents Tactus, an open-vocabulary tactile recognition model that maps low-cost pressure-array data to text embeddings, matching or exceeding a supervised closed-set CNN baseline on the STAG benchmark with only 187 training recordings and no classifier head.
DyPES-VLA is a cross-embodiment VLA model that learns shared dynamics priors via future prediction and uses an embodiment-specific Mixture-of-Experts action head to control robots in their native action spaces, achieving state-of-the-art results on LIBERO, RoboCasa, and RoboTwin benchmarks.
Xiaomi open-sourced Xiaomi-Robotics-1, an embodied AI foundation model pretrained on over 100,000 hours of UMI data and post-trained on 10,000+ hours of cross-embodiment data. The release includes the full real-robot post-training and deployment pipeline, aiming to challenge proprietary robotics models from Figure AI and Tesla.
Xiaomi released XR-1, a robot foundation model trained on over 100K hours of real-world manipulation trajectories. Built on Qwen3-VL and a Diffusion Transformer, it enables out-of-the-box mobile manipulation in unseen environments.
Tesla tweets about the relationship between AI software ('brain') and hardware/robotics ('body'), hinting at ongoing developments.
TechCrunch Disrupt 2026 announces a new Real World AI Stage focusing on AI in the physical world, featuring robots, automated factories, and de-extinction, with speakers from Shield AI, Colossal Biosciences, FieldAI, and more, October 13-15 in San Francisco.
Xiaomi announces Robotics-1, a new robot foundation model with a VLA architecture trained on 100K+ hours of real-world robot data, featuring cross-embodiment generalization and fast adaptation to new tasks.