The Unitree G1 humanoid robot recently completed a complex driving test under the control of software developed by Symbiosis Robotics. Rather than relying on traditional programming, the robot uses an end-to-end artificial intelligence model called Direct Perception Control. This system processes visual inputs and translates them directly into precise physical actions, allowing the machine to steer with its hands and operate the pedals with its right foot.
To achieve this level of whole-body coordination, developers trained the model on over 15,000 hours of diverse movement data. The go-kart demonstration requires the hardware to maintain physical balance while responding to the sudden forces of acceleration, braking, and cornering.
What other physically demanding tasks will end-to-end AI models enable these robots to master?
Media: VaderResearch on X
To achieve this level of whole-body coordination, developers trained the model on over 15,000 hours of diverse movement data. The go-kart demonstration requires the hardware to maintain physical balance while responding to the sudden forces of acceleration, braking, and cornering.
What other physically demanding tasks will end-to-end AI models enable these robots to master?
Media: VaderResearch on X
This website uses cookies
Comments 0
Be the first to comment!