Create robots that work in collaboration with humans
Advanced intelligent processing including object recognition, situation recognition, action planning, and action control will be required for the robotics-centered products that are expected to play a role in Sonyâs future business. Therefore, we are developing a robotics framework to support recognition processing and action planning. At the same time, we are working on hypothesizing an architecture for advanced intelligent processing and developing a demonstration system that uses our newly developed framework.
Controlling the whole-body cooperative motions of a robot requires creating a mathematical model of the entire mechanism to calculate the driving force of each joint involved in the individual motions. The computational technique for this is our unique generalized inverse dynamics (GID) methodology. Controlling the forces of all the joints using GID makes it possible to have any given part of the body move in many different ways and change the compliance as desired, with various constraints. This enables a variety of motions ranging from precise and rigid point-to-point tracking motions to loose and flexible motions to absorb collisions with the surrounding environment.
The virtualized actuator (VA) applies the joint forces calculated by GID so that it follows a nominal mathematical model while canceling out the actual friction and inertia. While various types of torque-controlled actuators exist, this virtualized actuator with Sonyâs unique patented technology enables the ideal response based on the mathematical model and delivers state-of-the-art stability. Moreover, while general torque sensors are susceptible to the state of motion and changes in the temperature, we minimize the impact of these factors by improving the mechanical design and electrical circuitry.
Robots perform movement and work by coordinating the external force generated by contact with their environment through their various parts (such as legs and arms) and the joint force that they themselves generate. Multi-contact Stabilization Control keeps the external force obtained from the environment at an appropriate value so the momentum of the entire system does not diverge while the contact area changes from moment to moment. By this principle, the legged robot can keep walking stably even if the ground is uneven.