What did you find out nobody did before?
Our hypothesis for this work was that so-called "explorer" agents can improve real-time motion planning for complex systems like bimanual robots or humanoids. These virtual agents explore the often fast-changing and dynamic environment for continually better trajectory predictions at a global level instead of a single path from the start to the goal. This was demonstrated through extensive simulation studies as well as real robot experiments. More technically, this work is the first instance of reactive and constrained planning for dual-arm systems. It can generalize to new scenarios and escape „local minima“, an inherent disadvantage for reactive methods. Local minima are points in the workspace where the system can get trapped due to the employed modeling strategy. It also can handle other tasks in the system like joint limits avoidance, such that physical limits of the actuators present in each joint are respected, or placing a held tray on a table or handling objects in a human-like fashion with a sense of touch.
Thanks to virtual agents: Service robots become more flexible
What are the unique challenges with two arms instead of one?
The main challenges with dual arms are twofold. First, at the operational space level – the reachable workspace of the robot where the task is defined – bi-manual tasks require spatial and temporal coordination of the hand as the end-effector at all times. And, it has to adhere to closed kinematic chain geometric constraints. This gets even more difficult with dynamic obstacles. Because many objectives have to be fulfilled at the same time, the workspace is now reduced. This has an impact on the control of the overall system. Second, issues at the joint control level need to be addressed. That is, actuators' motion control must be satisfied at each step while planning in cluttered spaces. Otherwise, the system gets stuck.
Where do you see practical scenarios for your research?
This research applies to humanoid pick-and-place scenarios in fast-changing environments where you would need to pick objects and place them on a table or a stacking location. Also, in a service robotics setting like Geriatronics where a helper robot like Garmi has to serve a tray with food from the kitchen to the bedside table. Our proposed integrated planning and control solution tackles geometric uncertainties and unwanted or desired contacts on the load.
Further information
- Video
- Publication: Predictive Multi-Agent-Based Planning and Landing Controller for Reactive Dual-Arm Manipulation; R. Laha, M. Becker, J. Vorndamme, J. Vrabel, L. F. C. Figueredo, M. A. Müller, S. Haddadin; IEEE Transactions on Robotics.