Is This the Future of Robotics? Unitree’s G1 Moves with Human-Like Precision

• Unitree’s G1 robot exhibits agility befitting human reflexes as it performs diverse stunts such as side flips, kick-ups, and moves mimicking combat. 
• G1 – Powered by progressive AI training and hardware, it denotes the new era of phenomenal anthropomorphic robotics, which shows immense real-world adaptive use.

Unitree’s newest footage from the testing grounds shows some rapid progress in humanoids, illustrating advanced innovations in robotics. 

The small G1 robot from the Chinese company does spectacularly to perform the very first kick-up, exhibiting a remarkable degree of agility and showcasing advanced capabilities of the system.

The company has been further continuously updating the algorithms of the robot so it can learn and execute more complex movements with high-grade precision and control.

Recently, Unitree shared footage of the G1 executing a side flip in a stand-up position. This is more than what the firm’s full-size humanoid avatar, the H1, was able to do in terms of standing backflips in March 2024.

Robotics Precision Mastered

Recently, in a video, Unitree’s G1 rocked the show with great agility and balance. It executed a combat sweep kick with its trainer and demonstrated some competitive Tai Chi movements. 

The robot was also able to upright itself from the face-up position with a bag in the way. In another test, while walking, G1 was kicked from behind and instinctively counted on maintaining its balance utilising its legs, hands, and body — almost mimicking a human-like reflex.

The demonstration really displayed the advanced movement and adaptability of the G1 and marks a significant advance in humanoid robotics.

Training takes place in a virtual prototype by harnessing Nvidia’s Isaac Simulator, where the robot learns complex behaviours before the physical deployment. Digital twins of the humanoid robot are made using motion capture and video data to imitate human behaviour. Reinforcement learning, in Unitree’s words, allows for fine-tuning of these behaviours within the simulator. In a process, they have coined Sim2Real, the virtual robot employs learnt skills onto the actual robot for real-time behaviour.

On the hardware side, it features 23 DoF via G1 to ensure much-improved stability and coordination. Standing 4.33 feet (1.32 m) tall and weighing 77 lbs (35 kg), it is designed to fold into a narrow pocket that allows it to be compactly stored down to 27 x 17.7 x 11.8 inches.

This advanced system has a three-dimensional LiDAR, a RealSense depth camera, and a noise-cancellation microphone array for voice command recognition. 

The G1 is powered by a 9,000-mAh quick-release battery that gives a maximum of two hours of run time with a fast-swapping solution. An 8-core high-performance CPU controls all the joints of the robot, hence allowing it to move fast and walk at a speed of about 6.5 feet per second (4.5 mph).

Open-Source Robotics

An open-source full-body joint-movement dataset developed by Unitree to facilitate motion control and coordination is available for humanoid robots. This dataset is valid for all of Unitree’s humanoid models – including the H1, H1-2, and G1 – enhancements for fluidity in motion, and allows dancing-type motions or repositioning dynamically in a human fashion.

The development includes the LAFAN1 motion capture technology, with an advanced redirection algorithm for optimised motion planning. This algorithm applies interactive mesh processing and inverse kinematics while considering constraints on posture, joint articulation, and velocity parameters.

By making the dataset public, Unitree hopes more innovation will follow and that real-world applications for humanoid robotics will go further.

Unitree’s humanoid models are continuously setting new benchmarks for the industry, with the H1 in March 2024 having attained a Guinness World Record with a running speed of 7.38 mph (3.3 m/s). The G1 is a compact and cost-effective platform for research purposes, aiding in the accessibility of the advanced robotics field, both in academia as well as the commercial sector.