Single assembly sequence to flexible assembly plan by Autonomous Constraint Generation

Abstract

The factory of the future is steering away from conventional assembly line production with sequential conveyor technology, towards flexible assembly lines, where products dynamically move between work-cells. Flexible assembly lines are significantly more complex to plan compared to sequential lines. Therefore there is an increased need for autonomously generating flexible robot-centered assembly plans. The novel Autonomous Constraint Generation (ACG) method presented here will generate a dynamic assembly plan starting from an initial assembly sequence , which is easier to program. Using a physics simulator, variations of the work-cell configurations from the initial sequence are evaluated and assembly constraints are autonomously deduced. Based on that the method can generate a complete assembly graph that is specific to the robot and work-cell in which it was initially programmed, taking into account both part and robot collisions. A major advantage is that it scales only linearly with the number of parts in the assembly. The method is compared to previous research by applying it to the Cranfield Benchmark problem. Results show a 93% reduction in planning time compared to using Reinforcement Learning Search. Furthermore, it is more accurate compared to generating the assembly graph from human interaction. Finally, applying the method to a real life industrial use case proves that a valid assembly graph is generated within reasonable time for industry. • ACG method autonomously generates all assembly sequences based on initial sequence. • Physics simulator evaluates deviations from initial sequence to deduce constraints. • Execution time scales only linearly with number of parts, states and skills. • Compared to previous research ACG is 93% faster and more accurate. • Industrial use case shows that ACG finds result within reasonable time for industry.