Grounded within a manufacturing context, the QAP team focuses on precision execution, creative task planning and flexible autonomy for manufacturing systems. We want to break the traditional paradigm of hyper-calibrated, fully-programmed manufacturing systems by marrying knowledge-based algorithms with data-driven techniques for transferring solutions across the manufacturing space. Furthermore, as the world moves towards Industry 4.0, and generally robots being ubiquitous everywhere, we need to have more advanced, grounded theories of Human-Robot Interaction with a focus on communication for coordination. This project serves as a testbed for such experiments, as well.
Ongoing Research themes:
- Vision for Manufacturing: Traditional knowledge-based estimation algorithms provide rigid guarantees which a real-world system requires, however, modern deep learning in vision lends a considerable edge in terms of execution run time. We are exploring hybrid architectures that can build on the best parts of these techniques.
- Task planning for high-mix, low-volume contexts: Usually manufacturing systems are calibrated to manufacture fixed combinations of fixed parts for efficiency. We are working on abstracting and rehauling the planning process such that we can:
- Learn from new examples to build novel combinations
- Transfer planning knowledge and processes from known tasks to similar new combinations
- Human-robot Interaction for Efficient Cooperation: In a similar vein to the above, robots will soon be “cage-free” (as ethics demand) in the workplaces. This requires a more dynamic understanding of how HRI and multi-agent cooperation changes between nominal and boundary conditions. Our current work looks at grounding the known theories of HRI and communication within the tangible context of manufacturing and extending them for fluid cooperation.
- Aayush Naik
- Hongyi Ling
- Jiaming Hu
- Priyam Parashar