Python package for learning safe robot trajectories released
Fast robot movements are essential for industrial production processes in order to achieve short cycle times and thus a high manufacturing throughput. If motion sequences are to be optimized using learning techniques, it must be ensured that neither the robots nor their environment are damaged during the exploration process. The Python package 'safemotions' (Preprint / Code / pip install safemotions) developed at IPR enables the learning of collision-free robot motions without exceeding kinematic or dynamic joint limits.
The method for complying with kinematic joint constraints (position, velocity, acceleration, jerk) is based on the publication "Learning Robot Trajectories subject to Kinematic Joint Constraints" (Preprint / Code), which was presented at this year's IEEE International Conference on Robotics and Automation (ICRA):
The 'safemotions' package also prevents collisions and torque limit violations by ensuring the existence of an alternative safe trajectory at all times. In addition to the source code needed to learn safe movements, the package also contains pre-trained neural networks for various reaching tasks with up to three robots (21 degrees of freedom).
Some exemplary motions are shown in the following video: