HARDWARE-ASSISTED RESEARCH PLATFORM FOR TOPOGRAPHICAL ROBOTIC CONTROL WITH INFINITE DOF
Abstract
This DURIP proposal aims at constructing a versatile hardware-assisted research platform dedicated to studying how to develop and implement alternative computational architectures and algorithms for next generation artificial intelligence (AI) that entails a huge quantum leap in our computing power and energy efficiency. Heavily relying on linearizing passive dynamics and nearly ideal sensing", today~s autonomous robotic learning often exhibits striking ~fragility and brittleness~ in unstructured and rapidly changing envir""onments. In sharp contrast, this proposed computing research is designed and implemented through scientifically studying how octopus"" ~thinks~ and ~reacts~, especially about how octopus uses a relatively small central brain to integrate a huge amount of visual and"" tactile information from the large optic lobes and the peripheral nervous system of the arms and, more importantly, issues commands"" to lower motor centers controlling the elaborated neuromuscular system of the arms, through the powerful framework of hierarchical" and stochastic-based deep learning that enable real-time data fusion and exploitation.The planned research areas will include 1)" developing autonomous learning algorithms and hardware for dynamical systems exhibiting nonlinearity, memory, and potentially infin""ite dimensionality, 2) investigating how to actively control highly distributed sensing and actuation that can enable optimal non-eq""uilibrium environmental energetic and information transfer to generate predictive models, and 3) studying how to effectively represe""nt uncertainty and decision time scales to enable real-time model inference, reinforcement learning, model-predictive control. All t"hese planned research activities will revolve around designing and implementing a hierarchical information processing computing hard"ware and software platform augmented by strong coupling between highly distributed sensing and actuation, which can ultimately culmi""nate with effectively controlling a 1000-DOF hyper-redundant robotic arm with extremely high computing performance, ultra-low energy"" consumption, and superior error resilience.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 23, 2018
- Source ID
- N000141812088
Entities
People
- Mingjie Lin
Organizations
- Office of Naval Research
- United States Navy
- University of Central Florida Board of Trustees