Rhoda AI Robot Platform
The Rhoda AI Robot Platform is a general-purpose mobile manipulator featuring a humanoid upper body (dual 7-DOF arms with dexterous hands) on a wheeled base. Developed by Rhoda AI (Palo Alto), it uses the FutureVision AI system trained on internet-scale video. It supports 25kg rated / 40kg peak payload and targets industrial automation tasks.
Primary use case: General-purpose industrial manipulation, warehouse automation, and factory operations
Payload
25 kg
Specifications
Physical
Locomotion
wheeled
Manipulation
Hand Type
gripper
Payload
25 kg
Max Lift/Arm
25 kg
Power & Autonomy
Autonomy Level
fully_autonomous
AI & Compute
AI Capabilities
vision, planning, manipulation, learning
Commercial
Availability
coming_soon
Primary Use Case
General-purpose industrial manipulation, warehouse automation, and factory operations
Related news
More $$$ for robots: Matter Venture Partners closes $450M Fund II to back robotics startups
HD Hyundai Robotics Unveils AI Robot With Human-Like Arms - 조선일보
HD Hyundai Robotics unveiled an AI robot described as having human-like arms. No further details, specifications, funding, investors, or people are provided in the supplied text.
Rhoda AI Shows Robots Recovering From Factory Edge Cases in Schrep Visit
During a Field Trips with Schrep visit, Rhoda AI demonstrated robots recovering from accidental grasps, extra rubbish, trapped paper and repositioned automotive radar units while unpacking a 10 kg box of bearings and moving parts through inspection. Its Direct Video-Action architecture predicts future video frames and uses a separate inverse-dynamics model to translate them into actions, with 10–20 hours of task-specific teleoperation data required after extensive video pretraining. Rhoda AI is targeting paid production deployments before the end of 2026, but the demonstrations did not establish sustained factory reliability.
Introducing Instant Robot Learning
Machenta, a frontier physical AI company and Imperial College London spinout, announces a robot foundation model that learns new tasks from a single demonstration with no further training. The Vision-Demo-Action (VDA) model is trained entirely in simulation with zero real-world training data, making learning how to learn the optimisation target rather than task memorisation. The same pre-trained model is generalist, task-agnostic and cross-embodiment: it runs across six robot embodiments with different cameras and grippers, supports four teaching modes (leader-follower, VR controller, kinaesthetic teaching, and bare hands), and extends to humanoid hands. Demonstrations include instant spatial generalisation, reactivity and recovery, compositional object generalisation, demo stitching, and a one-shot one-person factory running three robots on one model. Also announced: a $40M seed round co-led by Creandum and Union Square Ventures, the largest seed raise by a UK robotics startup. The founding team traces to six years of one-shot robot learning research at Imperial's Robot Learning Lab (Implicit Graph Alignment, CoRL 2023; Instant Policy, ICLR 2025).