Robotic manipulation at the micro- and nano-scale can fundamentally transform disease treatment and the assembly of small objects. The goal of this project was to precisely deliver materials and assemble structures from the bottom-up. This precision manipulation must be coupled with a large population of manipulators to enable rapid progress. The potential impact is broad: large populations of micro-manipulators could provide targeted therapy, perform minimally invasive surgery, and engineer tissue. However, the small size of micro- and nano-robots severely limits their computation, sensing, and communication capabilities. This project designed new techniques for centralized control under the constraint that every robot receives exactly the same input commands.

This proposal introduced massive manipulation to solve this problem. Massive manipulation uses a shared input to drive large populations of robots to arbitrary goal states. The unifying theme is using obstacles to efficiently control the shape, arrangement, and position of the swarm. Fortunately, in vivo environments are rich in obstacles, and artificial workspaces can be engineered to exploit these techniques. Two broad techniques are applied: 1.) Designing feedback controllers for controlling swarm configuration, manipulation through obstacles, and multi-part assembly. These controllers learned from human-user data collected from the citizen-science website. 2.) Algorithmic techniques for massively-parallel assembly, efficient aggregation, and reliable coverage. Control laws and algorithms were validated with analytical models, extensive simulations, and scaled hardware experiments using 100 or more kilobot robots. These scaled hardware experiments enable rapid reconfiguration and emulating a variety of micro-scale dynamic models provided by collaborators.

Over the course of this award, we built algorithms that, given a desired shape of particles, would design a 'factory' arrangement of obstacles that, when populated with particles, would build identical copies of the desired shape every two global inputs. These global inputs could be either physically tiling the workspace in two directions, or global magnetic forces. 

We also designed automatic controllers to command 100 simple robots to work as a coordinated swarm. The robots were steered by turning on lights situated around the table they were moving on.  We use the global input of which light is turned on to make the swarm push the objects 100x the size of one robot through mazes to desired goal location.

We built algorithms to rapidly gather particles, used particles to explore unknown workspaces, designed algorithms to reconfigure robots into desired shapes and patterns, and designed modular robotic cubes with embedded magnets that can be actuated in unison by an external magnetic field to move, assemble, and disassemble.  Collisions with boundaries enable reconfiguring the cubes, embedded magnets encode which cubes can bond. We used reinforcement learning to rapidly gather particles with potential applications in targeted drug delivery.

This award supported three PhD students who are now engineers at US companies, and partially supported 10 Master of Science students as they wrote their theses. Papers were presented at IROS, ICRA, CASE and in the journals Robotics and Automation Letters, Transactions on Automation Science and Engineering, and in Transactions on Robotics.

Last Modified: 10/05/2022
Modified by: Aaron T Becker