Best Use of AI Finalists

Borg Robotics 

Borg Robotics is an AI robotics company pioneering autonomous solutions for industrial environments, with its flagship humanoid, Borg 01, showcasing advanced AI capabilities in real-world tasks. The Borg 01 is a modular humanoid robot designed for manufacturing and warehouse operations, capable of switching between walking and wheeled modes while autonomously carrying out logistics tasks. Embedded neural networks and onboard AI algorithms drive its perception, decision-making, and task adaptation, enabling it to recognize its surroundings, adjust to changing conditions, and execute tasks. 

Diligent Robotics  

Diligent Robotics’ humanoid robot Moxi uses advanced AI to operate in hospital settings, adapting to new environments and workflows. Moxi seamlessly navigates busy corridors, opens doors and elevators, and avoids people and obstacles, all while behaving in ways that feel natural and comfortable to staff and patients. Powered by artificial intelligence, Moxi has already saved hospital staff nearly 600,000 hours and successfully completed over 1 million tasks. 

Galbot 

Galbot’s G1 humanoid robot made headlines by autonomously running a convenience store in Beijing – greeting customers, fetching goods, and handling sales without any human teleoperation. This 173 cm, 85 kg humanoid uses multi-modal AI (vision, touch, and large language/vision models nicknamed “GroceryVLA” and “GraspVLA”) to perceive its surroundings and manipulate objects with dexterity. With voice recognition and safe human-robot interaction capabilities, Galbot demonstrates advanced AI in humanoids by blending autonomous navigation, object recognition, and natural interaction. 

Kepler Robotics 

Kepler Robotics’ K2 humanoid models bring cutting-edge AI to life, tackling both industrial and everyday tasks with impressive precision. Recently showcased at SAIC-GM’s factory, the K2 wowed observers by performing delicate, high-precision operations with its adaptable, human-like grasping. Thanks to advanced object recognition and scene understanding, it effortlessly identifies parts on an assembly line and navigates even the most cluttered environments, making it a true game-changer for modern automation. 

Macco Robotics 

Macco Robotics’ humanoid, KIME the bartender, leverages AI in several ways to ensure efficient service: it uses computer vision to precisely measure and monitor liquid levels when pouring beverages, ensuring consistent results with every drink. It’s also equipped with facial recognition and a touch-screen interface, allowing it to recognize returning customers or process cashless payments (for example, identifying a patron’s age for alcohol service.) KIME not only improves efficiency but also provides an engaging, tech-savvy experience for customers. 

PAL Robotics 

Their flagship humanoid TALOS is a full-size bipedal robot designed as an open platform (fully ROS-enabled) for researchers to develop custom AI. TALOS features advanced whole-body torque control and has been used in projects (e.g. with Airbus) to explore using AI-driven humanoids for industrial assembly. PAL Robotics also experiments with cutting-edge techniques like reinforcement learning: for instance, their new Kangaroo biped uses model-based control combined with RL-trained controllers to achieve agile, robust locomotion and jumping, learned in simulation with GPU acceleration. 

Pollen Robotics 

Pollen Robotics’ humanoid robot Reachy 2 is an open-source humanoid torso with two arms and an expressive head, purpose-built for research in manipulation and social interactionrobotsguide.com. Reachy is modular and fully programmable in Python/ROS, which allows developers to plug in custom AI algorithms for vision, planning, or dialogue. Reachy has been shown performing teleoperated tasks and semi-autonomous behaviors: for example, an earlier version of Reachy won 2nd place in the ANA Avatar X-Prize by remotely executing difficult tasks like using a drill. 

Robot Era 

Robot Era unveiled ERA-42, an end-to-end large, embodied AI model integrated with its humanoids. ERA-42 is essentially a massive neural network trained for robot control; paired with Robot Era’s five-fingered XHand1 dexterous hand, it allows the robot to perform over 100 intricate tasks (using tools, assembling objects, etc.) with human-like precision. Notably, ERA-42 learns new tasks in just a couple of hours from minimal data and generalizes its knowledge to different scenarios, eliminating the need for painstaking manual programming. 

Tesla 

Tesla’s Optimus has been seen performing a variety of everyday tasks this last year and has helped achieve this by training a single large neural network that controls the robot’s every move across all these tasks. The robot’s “brain” is taught via imitation learning: Tesla fed Optimus with first-person video data of humans performing the target tasks, allowing the AI to learn the required actions and tool use directly from human examples. According to Tesla’s Optimus team, this approach lets the robot quickly learn new skills from human demonstration videos and continually refine its abilities. 

UBTECH Robotics 

UBTECH’s humanoid robots apply AI across perception, interaction, control, and cloud services to support education, entertainment, and light commercial tasks. UBTECH robotics Walker S2 is powered by a dual-loop AI system that includes BrainNet 2.0 & Co-Agent, Walker S2 achieves progressive evolution in both single-agent autonomy and multi-robot collaboration. BrainNet 2.0 is UBTECH’s software framework for multi-humanoid coordination and swarm intelligence, designed to connect cloud and on-device inference nodes into a hierarchical “super brain”. At the China Pavilion of Expo 2025 Osaka, UBTECH robotics showcased its Walker C, the first embodied AI ”Tour Guide” welcoming global visitors.