Researchers propose ROSA, a robotics foundation model serving system designed for robot factories that moves beyond single-robot, edge-computing assumptions. The system utilizes shared GPU-pool serving to allow fleets of robots to access server-class GPUs over the network.
Calibra v0.7.1 introduces a new Dataset Integrity stage that runs before quality and coverage analysis to help users verify if their robotics datasets are trustworthy.
Google DeepMind has released Gemini Robotics 2, an intelligence layer comprising three models designed to enable whole-body control, five-finger dexterity, and multi-robot collaboration. The release includes a Vision-Language-Action (VLA) model, an embodied reasoning VLM, and an on-device VLA, moving beyond previous table-top manipulation capabilities.
Google has introduced Gemini Robotics 2, an intelligence layer designed to provide adaptable robots with intelligent whole-body control, advanced dexterity, and the ability to collaborate in teams. This update expands physical AI capabilities beyond narrow, repetitive tasks by enabling robots to reason through movements and adapt to new bodies quickly.
Google has launched Gemini Robotics ER 2, a new embodied reasoning model designed to help robots think fast and plan multi-step tasks in real-time. The update introduces significant upgrades over the previous ER 1.6 version, including continuous video-based progress tracking, precision moment-finding, and native multi-robot collaboration capabilities.
Calibra is an open-source toolkit designed to audit robot datasets and identify quality issues before training. It provides tools for building quality-aware coresets and estimating training outcomes.
Calibra is an open-source toolkit designed to help researchers audit robot datasets and identify quality issues before training policies. The first public release includes a Robot Dataset Health Check interactive Space for auditing LeRobot datasets and a benchmark of 30 public LeRobot datasets with health scores.
The University of Pittsburgh’s Human Engineering Research Laboratories (HERL) is leading the Robotic Assistive Mobility and Manipulation Platform Providing Independence for People with Disabilities (RAMMP), a project backed by up to $41.5 million in funding from ARPA-H. The initiative aims to create safer, more adaptable assistive mobility solutions for wheelchair users by integrating advanced robotics with Meta’s open-source AI vision models.
NVIDIA has introduced Cosmos-H-Dreams, a real-time, action-conditioned generative simulator for surgical robotics that distills the capabilities of its Cosmos-H-Surgical-Simulator into a causal student model. Served through NVIDIA's FlashDreams accelerated streaming-inference library, the system enables interactive control by a person or learned policy in a closed loop.
Black Forest Labs has launched FLUX 3, a multimodal flow model family that includes video generation capabilities with native audio. The release also introduces FLUX-mimic, a variant designed for video-action robotics applications.
This article provides an overview of the current state of simulation for Physical AI, explaining how simulation bridges the data gap for robotics by enabling scalable, photorealistic training environments. It outlines a three-computer paradigm (training, simulation, on-robot) and categorizes key simulation engines based on their specific capabilities and use cases.
NVIDIA has released Cosmos 3 Edge, a 4-billion-parameter open world model designed to run on-device for robots and vision AI agents. Released July 20 on Hugging Face, it enables systems to understand surroundings, reason in real time, and generate robot actions locally without cloud dependency.
Pollen Robotics has released Grabette, an open-source handheld device designed to record real-world robot manipulation data without requiring a robot or teleoperation rig. The system uses a fisheye camera and an RGBD depth camera to capture 6-DoF trajectories, allowing users to generate clean datasets that can be processed via a browser-based dashboard and shared on the Hugging Face Hub.
NVIDIA has released Cosmos 3 Edge, a 4-billion-parameter open world model designed to deliver data center-level performance on memory-constrained edge systems. The model enables robots and vision AI agents to understand their surroundings, reason in real time, and generate actions directly on devices such as NVIDIA Jetson modules.
Mistral AI has released Robostral Navigate, its first model designed for embodied navigation using only a single RGB camera. The 8B-parameter model accepts plain-language instructions and image inputs to guide robots through complex environments like offices and outdoor settings.
Researchers introduce RoboTTT, a training recipe that extends visuomotor context for robot foundation models to 8,000 timesteps without increasing inference latency. The method integrates Test-Time Training into Vision-Language-Action policies by using fast weights updated via gradient descent during both training and inference.
NVIDIA introduces RoboTTT, a robot foundation model and training recipe that extends visuomotor context to 8,000 timesteps without increasing inference latency. By integrating Test-Time Training into Vision-Language-Action policies, the system compresses history into weight space using fast weights updated by gradient descent.
Xiaomi has introduced Xiaomi-Robotics-U0, a 38-billion-parameter multimodal autoregressive model designed for unified embodied synthesis. The model treats embodied generation as an extension of foundation image and video generation, jointly optimizing text-to-image generation, image editing, embodied scene generation, embodied transfer, and embodied video generation.
This paper introduces PAC-ACT, a reinforcement-learning post-training framework designed to enhance pretrained Action Chunking Transformer (ACT) policies for industrial precision-contact tasks. The method reformulates policy optimization at the chunk level and utilizes an ACT-transferred actor-critic architecture with a hybrid behavior-prior constraint.
NVIDIA addresses the challenge of rigorously evaluating robotics foundation models, which can now follow natural language instructions to manipulate objects. The company identifies evaluation as a critical unsolved problem and presents a new method to address these issues.