Panasonic Develops New Robotic Motion Teaching Technology

Motion Teaching is Traditionally Difficult in Some Situations

Robotic programming can be an expensive, time-consuming task that continues to grow in importance as industry increases the amount of robotic technology used in manufacturing and logistics. In the past, robotic programming consisted of a trained robot programmer who would manually move the robot through points in space and create code that the robot would follow without deviation.

This method of programming works well in environments that rarely, if ever, experience any change, such as hard robotic systems that exist behind safety fences. But what about situations that involve collaborative work with humans or other environments that are often presented with changing parameters? These situations are typically quite challenging, and Panasonic Holdings Corporation has developed new technology to help in these types of situations.

 

The motion of robots in coordination with tasks, people, and other robots is prone to variations in the environment and requires some degree of flexibility.

 

Task Segmentation for Faster Collaborative Programming

Panasonic Holdings Corporation has created new technology capable of teaching robotic motions in differing environments, maintaining a high level of performance without sacrificing robotic safety and accuracy. It has been traditionally difficult to keep a high level of accuracy in situations where the robot is in contact with people and other objects. The new technology treats the robot in a different context than before, providing the ability to program the robot on the assumption that the robot is behaving in its environment flexibly, much like the motion of a spring in a dynamic system.

In most robot programming, constant power is sent to the motors, causing them to operate with a certain rigidity (or impedance) against other objects. If the robot runs into an obstacle, it is more likely to damage that item rather than being flexible enough to come to a stop before its pre-determined destination. Think of a robot opening a door, the robot gripper navigates to the door handle, presses down, and then pulls. But what if the door is slightly ajar? The robot will contact the door and either push it or come to a stop, sensing a collision. If the system is designed with more flexibility and lower impedance, the robot could sense that it has contacted the door handle a bit early and simply proceed with its objective without landing on the predetermined point with sub-millimeter precision.

This new system is able to learn a more advanced impedance control in order to perform work tasks. It works by using variable impedance for work tasks as well as using traditional impedance control for worker and robotic safety. The system uses multi-objective Bayesian optimization and motions taught through human demonstration. It has been recognized by the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2023, and will be presented at the plenary session to be held in Detroit, Michigan in October.

 

How Does the Technology Work?

The technology uses advanced impedance control by establishing the robot as a so-called ‘spring’ within the system through impedance gain. By reading the impedance in the robot at any given time, the system is able to identify changes in the robot's environment. This is how collaborative robots have worked in the past, but there are limitations with a simplistic impedance control algorithm.

Simplistic impedance control is faced with the problem of setting an impedance gain that maximizes both safety and accuracy, usually as one increases, the other decreases.

 

Separated impedance control for robotic door opening, where three clearly different tasks are defined, with different impedances defined for each task. Image used courtesy of Panasonic Holdings Company

 

Panasonic Holdings Company has managed to increase the accuracy of the robots without sacrificing safety. It is possible through the use of segmented tasks, where the robot separates tasks into different segments, and an impedance gain is set for each of the segments. The segments are created through the use of prior knowledge, then the robot is run through the tasks several times to increase the accuracy of the system.

Once it knows approximate values, it can then operate within normal parameters and set the impedance gain to reflect the portion of each task as it completes them. If the robot sees an impedance gain outside of its normal parameters, it then knows that something has changed and can take appropriate action. For example, if a robot comes into contact with a worker, the impedance will increase because more energy will be required to continue the robot's movement.

 

Advantages of Motion Programming

Many of the finer algorithm details of motion programming will never be faced by many engineers. However, understanding the concepts that drive motion can help to build safer, faster, and more accurate systems that are less prone to failures in the future.