Real-time path planning for automating optical tweezers based particle transport operations

摘要: Title of Dissertation: REAL-TIME PATH PLANNING FOR AUTOMATING OPTICAL TWEEZERS BASED PARTICLE TRANSPORT OPERATIONS Ashis Gopal Banerjee, Doctor Philosophy, 2009 Directed by: Professor Satyandra K. Gupta Department Mechanical Engineering Optical tweezers (OT) have been developed to successfully trap, orient, and transport micro nano scale components many different sizes shapes in a fluid medium. They can be viewed as robots made out light. Components simply released from optical traps by switching off laser beams. By utilizing the principle time sharing or holograms, multiple perform several operations parallel. These characteristics make very promising technology for creating directed assemblies. In infra-red regime, they are useful large number biological applications well. This dissertation explores problem real-time path planning autonomous OT based operations. Such pose interesting challenges environment is uncertain dynamic due random Brownian motion particles noise imaging measurements. Silica microspheres having diameters between (1-20) μm selected model components. Offline simulations performed gather trapping probability data that serves measure trap strength reliability function relative position particle under consideration with respect focus, velocity. Simplified models generated using Gaussian Radial Basis Functions represent compact form. metamodels queried at run-time obtain estimated values accurately efficiently. Simple then utilized stochastic programming framework compute optimum locations velocities minimizes total, expected incorporating collision avoidance recovery steps. A discrete version an approximate partially observable Markov decision process algorithm, called QMDP_NLTDV developed. Real-time performance ensured pruning search space enhancing convergence rates introducing non-linear value function. The algorithm validated both simulator well physical holographic tweezer set-up. Successful runs show automated planner flexible, works reasonably crowded scenes, capable transporting specific given goal location avoiding collisions either circumventing other freely diffusing particles. technique individual within decoupled prioritized approach move simultaneously. An iterative bipartite graph matching also used assign target objects optimally. As case single transport, simulation some experiments validate multi-particle approach.