Reward AI Demonstrates Humanoid Robot Delivery Skills
The startup uses sensor gloves to train robots on human motor patterns for warehouse and cleaning tasks.
Updated on Oct. 31, 2026 in Robotics
Live Poll
Do you believe humanoid robots will soon reliably handle complex tasks in everyday settings?
Robotics startup Reward AI has released a demonstration video of its humanoid robot performing delivery and cleaning tasks. The robot operates using the OM-1 AI model, which the company trains by capturing human hand motions and grip pressure through sensor gloves.
Why it matters
The company uses human-in-the-loop sensor training to overcome the scarcity of high-quality robotics training data. This approach represents an alternative path to the high-scale virtual simulation methods being used by competitors to develop generalized autonomous movement.
Reward AI uses sensor-equipped gloves to record human grip force and trajectory for model training. This contrasts with Nvidia’s recent training approach using 4,096 simultaneous virtual environments.
The players
Reward AI
A United States robotics startup developing humanoid platforms using sensor-glove motion data.
A global technology conglomerate developing the Gemini Robotics 2 humanoid platform.
Nvidia
A semiconductor and software firm providing large-scale virtual simulation platforms for robotics training.
The details
The OM-1 model is trained by mapping human physical interactions onto the robot's action space. By recording nuanced hand motions via sensor gloves, the system aims to bypass the limitations of traditional programming. This research-stage method seeks to generate training data that captures the physical constraints and dexterity required for handling objects in unstructured environments.
Timeline
July 2026: Google unveiled its Gemini Robotics 2 humanoid.
October 23, 2026: Nvidia released demonstration footage of its MicroDuck robots.
October 27, 2026: Reward AI published its humanoid robot delivery video.
The Tech Race
The industry is currently divided between simulation-heavy training, such as Nvidia's 4,096-environment approach, and sensor-based human imitation. Developers are racing to reach the 80 percent performance threshold in new environments, a benchmark suggested by Unitree founder Wang Xinxing.
The humanoid robot remains a research-stage project and is not yet available for commercial or personal use. Future progress will be measured by whether these robots can transition from controlled demonstration environments to performing reliable labor in real-world settings.
The takeaway
The race to achieve reliable humanoid performance is currently split between high-volume virtual training and physical mimicry. Observers should track upcoming benchmark disclosures from these companies to see which training method successfully bridges the gap to real-world deployment.
Further reading
For broader trends in machine-driven movement, read more at Robotics.
Source note: This article includes information reported by 경향신문.
Live Poll
Do you believe humanoid robots will soon reliably handle complex tasks in everyday settings?
