Sanctuary AI Achieves Milestone with 21 DoF Robotic Hands Capable of In-Hand Manipulation

Sanctuary AI's dexterous robotic hands offer in-hand manipulation

The new robotic hand technology enables in-hand manipulation, unlocking the capability to perform high-value tasks requiring precise control. Sanctuary AI claims to have reached a milestone in robotics, advancing toward its goal of developing human-like intelligence in general-purpose robots. The company's dexterous robotic hands, featuring 21 degrees of freedom (DOF), now enable intricate in-hand manipulation. The advancement expands the potential for robots to perform complex industrial tasks requiring fine motor skills. The Vancouver-based AI and robotics company's dexterous robotic hands use miniaturized hydraulic valves, offering higher power density over cable and electromechanical systems. The innovative actuation enhances speed, strength, control, durability, impact resistance, and heat management.

Sanctuary AI's Phoenix Now Has a Robotic Hand with 21 DoF | AI News

The official video shows seamless transitions between human-like actions such as picking, reorienting, and placing objects. Vancouver-based AI and robotics company Sanctuary AI has announced another leap with its advanced robotic hands, which are capable of performing in-hand manipulation. This development enhances the dexterity of its general-purpose robots, paving the way for their use in intricate industrial tasks. The robotic hands feature 21 degrees of freedom (DoF) and employ miniaturised hydraulic valves for actuation, as mentioned on the official blog. This advancement enables seamless transitions between picking, reorienting, and placing objects -- tasks essential in industrial settings. The robotic hands are integrated into the company's general-purpose robot, Phoenix. This enhances the training of Sanctuary AI's control system, Carbon, by providing detailed behavioural data. According to the announcement from the company, this system delivers higher power density compared to traditional cable and electromechanical methods. It offers improved speed, strength, controllability, cycle life, impact resistance, and heat management. "Our ambition is to build a reliable system with human-level dexterity. We have also recently achieved a milestone of testing our hydraulic valve actuators over 2 billion cycles without signs of leakage or degradation," said James Wells, interim CEO and CCO at Sanctuary AI. The modular design of the system supports future collaborations with both humanoid and non-humanoid robots. "Demonstrating in-hand manipulation with a scalable and reliable system is a key milestone towards demonstrating the breadth and viability of capable general-purpose robots," Wells added. Sanctuary AI, established in 2018, claims that it aims to create robots with human-like intelligence. The company's robots are designed to address global labour challenges while promoting safety and efficiency in the workplace. In October of this year, Tesla also showcased the company's humanoid robot, Optimus, displaying 22 DoF, as it played rock, paper, and scissors with a fan at the 'We, Robot' event. Optimus has also recently mastered walking over rough, slippery terrain. Hydraulic systems provide significantly higher power density compared to electromechanical actuators, resulting in enhanced controllability, faster and stronger movements, and millimetre-level finger precision. Historically, integrating hydraulics into humanoid designs posed challenges due to the unavailability of off-the-shelf components. Sanctuary AI has addressed this by miniaturising actuators and valve systems to approximately the size of a coin. These components have undergone rigorous testing, completing over two billion cycles without leakage or degradation, thereby setting new standards for durability. The system operates at low pressure, comparable to consumer-grade hydraulic bicycle brakes, and utilises food-safe oil, ensuring safety and reliability. Additionally, it offers force estimation throughout the fingers, enhancing precision in manipulation tasks.