Vega

Vega, manufactured by Dexmate, represents a breakthrough in general-purpose mobile humanoid robots, launched in early 2025 and entering production with pre-orders at approximately $96,000. Designed primarily for industrial automation, R&D, manufacturing, logistics, and retail sectors, Vega addresses labor shortages through its versatile architecture combining high dexterity, payload capacity, and mobility. Unlike legged humanoids, Vega employs an omni-directional wheeled base for precise, efficient navigation at speeds up to 4 km/h, eliminating walking-related energy inefficiencies while maintaining a low profile for human-safe collaboration. The robot's core architecture features a highly configurable kinematic chain with 36+ degrees of freedom (DOF): a 3-DOF flexible head for panoramic spatial perception, 3-DOF adjustable torso enabling vertical extension from compact storage to a maximum reach of 2.2 meters (7'2"), and dual 13-DOF dexterous arms (7-DOF shoulder/elbow/wrist + 6-DOF five-fingered hands per arm). This setup supports bimanual manipulation of heavy objects up to 15 kg (33 lbs) per arm at any pose, ideal for assembly, inspection, picking, and high-reach tasks. The foldable torso and arms allow seamless transition between deployed and transport modes, weighing 135 kg overall with aluminum structural framing for durability and lightness. Propulsion relies on brushless motors paired with harmonic gear reductions for smooth, quiet, backlash-free motion. Sensing is comprehensive: multiple RGB/RGBD cameras for vision, LiDAR for SLAM/navigation, IMUs for odometry/orientation, ultrasonic sensors for obstacle avoidance, force/torque sensors on joints/end-effectors, and tactile touch sensors on hands. Safety features include fast self-collision avoidance, compliance control, and human-proximate operation logic. Compute architecture leverages NVIDIA Jetson AGX Orin (up to 64GB RAM, with Thor upgrade imminent) for edge AI processing, augmented by Intel x86 CPUs. Running Linux with an open ROS/Python stack, it offers pip-installable APIs, URDF/USD models for simulators like NVIDIA Isaac Gym/GR00T, and rich peripherals (4x USB 3.2, Ethernet, DisplayPort, 5V/12V rails). LLM integration enables natural-language interfaces for tasking and skill re-authoring, supporting generative AI pipelines for policy learning from demos/teleop (VR/exoskeleton compatible). No proprietary closed-source models are specified; emphasis is on developer-friendly extensibility. Real-world deployments include pilots with public companies in manufacturing/logistics/retail as of mid-2025, showcased at ProMat 2025, NVIDIA GTC, and ICRA events. Developed in under six months by a team from MIT, CMU, UCSD, NVIDIA, and TRI, Vega boasts 20+ hour runtime (up to 30+ hours light bimanual motion) on a high-capacity battery, with 3-month lead times. Its scalability supports fleet management, positioning it for warehouse automation, assembly lines, and research in dexterous manipulation. Early demos highlight barbell lifting, household chores, and precision handling, underscoring its transition from prototype to deployable platform amid the 2025 humanoid surge.