Great ideas are meant to leave the laboratory
Gentle’s capabalities stem from work by our dedicated engineering researchers who implement and develop cutting-edge Machine Learning and Autonomous Machine research, applied to practical applications.
Hardware
The first thing most people notice about Gentle machines is how many arms they have! By moving multiple arms in parallel, Gentle machines provide great accuracy and range of motion but keep the machine from needing to lift its own weight. This, in turn, allows the motors to move faster and smoother without gaining dangerous momentum as they move. It also allows the arms to be pushed and pulled easily in any direction you want without worrying about the joints restricting your movement.
Vision Core
High speed color videos are streamed from multiple viewpoints: from the robot base, from the gripper itself, and from a perspective camera at the edge of this machine. Video from the three cameras is combined to ensure the environment is mapped from the full reach of the machine down to the detailed location of the tool in its hand.
Motion Engine
The motion engine performs your task by coordinating hardware motions with the vision core’s map of the scene. The Gentle Mind motion engine is built from day one to learn from physical demonstrations by humans – not from hand-written code. This provides it a great ability to discover new ways to achieve tasks beyond simply following pre-designed motions. We give the Gentle Mind this autonomy to make some decisions, but always restrict the motions it can implement to stay within strict safety boundaries.
Spinal Network
Machines establish a robust and redundant remote-control link using low-latency 5G networks. Predicting future forces and motion across the network provides “transparent” movement and feeling across thousands of miles, the same way a phone call feels like it is in real-time. While in operation, the experience of each machine is accumulated, used to train better future behaviors, and then carefully erased (for complete user privacy).
Want to learn more? Click any of the papers below to download.
Pixel-to-Pose Estimator for Precise Robotic Tool Alignment
Teleoperated Hexapod Robot for Imitation Learning Task Training
Bowden Tube NiTi Actuators with Linear Parameter Varying Model and Sliding Mode Control
Linear Parameter Varying Modeling and Estimation of a SMA Wire Actuator
The Design of Optimal Lattice Structures Manufactured by Maypole Braiding
Dual Measurement Self-Sensing Technique of NiTi Actuators for Use in Robust Control
