Advancing impact-aware robot control for automation in logistics
Jari van Steen defended his PhD thesis at the Department of Mechanical Engineering on March 19th.
Labor scarcity is a rising problem in large parts of the world. The logistics sector, with applications like luggage handling at airports and parcel handling in warehouses, is no exception. Automation can be a solution to this problem in logistics, as robots can take over strenuous work such as picking up things and putting them somewhere else. However, not all human skills translate easily to robotics. A great example is the human skill of using intentional impacts to our advantage in activities like running, jumping, or kicking a ball. Jari van Steen shows in his PhD research how to exploit impacts in robotics in a human-like way. The results of this research can, among many other possible applications, speed up processes such as automated pick-and-place operations.
So far the focus has been on avoiding impacts in state-of-the-art robotics. That鈥檚 because impact-aware robotic manipulation comes with significant challenges. For example, impacts are tied to rapid jumps in velocities which are used to compute the actuator inputs. Naive control strategies can result in sudden unwanted jumps in the input signals which can lead to unpredictable robot behavior, including possible destabilization. This destabilization and the possible hardware damage and extended periods of downtime because of it are detrimental for any industrial use case. Jari van Steen proposes a novel control approach in his PhD thesis to remove input peaks while accurately performing industry-relevant tasks with impacts such as quick grasping of parcels using a dual-arm robotic setup.
Experimental validation
The research provides an experimental validation on different levels. Firstly, by showing the effectiveness of the proposed control approaches in avoiding input peaks against a set of baseline control approaches. Secondly, a comparison is made between impact avoidance and impact exploitation for an industry-relevant use case enabled by an integration of the proposed control approach with other impact-aware manipulation techniques. The comparison validated that exploiting impacts for a dual-arm pick-and-place use case is on average 16% faster than impact avoidance when the techniques shown in Jari van Steen鈥檚 thesis are used. This substantial increase in speed is a step towards making this technology become viable from a business perspective and enabling an expansion of robotic automation in logistics.
This research is part of the Horizon 2020 project I.AM., coordinated by the 果冻传媒 and involving the Eurotech Alliance partners EPFL and TUM.
Title of PhD thesis: . Promotors: Associate Prof. Alessandro Saccon and Prof. Nathan van de Wouw.