This study explores a real-world problem in industrial automation that is how to make industrial robots more energy sufficient. The area explored to achieve this is the use of machine learning to get energy optimized speed and acceleration of the robot while attempting to explore the trade off with time. Traditional robotic motion planning often relies on predefined rules, which usually may not effectively balance energy consumption and operational speed.

This study uses data from 100 pre-simulated robotic paths of a known pick and place task where each simulation has different speed and acceleration profiles to use as data for machine learning. The machine learnings that are used for this study is Random Forest and Neural Network models to predict how the factors of speed and acceleration affect both energy consumption and operation time. Once the models were trained, a numerical optimizer was used to find the best combination of speed and acceleration that can minimize energy usage. Using this data-driven approach, the adapted predicted optimized parameters which are 600 m/s speed and 55 m/s² acceleration were tested in Robot Studio.

This resulted in a 35% reduction in energy consumption and 48% of increase in cycle time with for a specific time weight of 0.07. The study directs to adaptive weighting in optimization, where the importance of execution time can dynamically adjust based on workload conditions. In addition, it also contributes to the advancement of smart robotics, paving the way for real-time optimization systems capable of responding intelligently to varying operational demands.