Research Areas
Motor Neuroprosthetics
The long-term goal of this study is to perform the basic science research required for developing the next generation of closed-loop brain machine interfaces that automatically extract sufficiently accurate and spatially resolved signatures and deliver meaningfully feedback directly to the brain. Our work addresses three pillars of neuroprosthetic design from a signal processing and neural engineering perspective: distributed sampling of neural activity directly from the cortex, optimizing the interpretation its information, and delivering feedback via biologically inspired models for generating electronic stimulation. The modeling and analysis research plan is part of a closed-loop system where the individual has the ability to interact, respond, and adapt to the neuroprosthetic device. This experimental paradigm opens many new questions about optimal neural decoding, techniques for neuronal sampling, and functional neurophysiology that were previously unable to be addressed with conventional single electrode, acute techniques.
See the publications section for details on all projects.
Co-Adaptive Brain-Machine Interfaces via Reinforcement Learning
Our laboratories are investigating closed-loop BMIs for reaching tasks with behaving animals interacting with a reinforcement learning (RL) algorithm to develop new treatments for the human condition. Here, we focus on the mechanisms of goal-based learning for motor BMIs through the control of a robotic arm. The approach is based on synergistic interaction between the user’s neural response and an intelligent agent that are both seeking to maximize their own reward. For the development of BMIs, the RL framework provides a mechanism of learning that is very similar to operant conditioning of biological organisms because the learner (patient) is not told what actions to take but must discover which actions yield the most reward by trying them. In this BMI experimental paradigm, one has access to the environment, the actions, the rewards, and also the brain signals, i.e. one can observe the spatio-temporal activation of brain states (indirectly related to the environment) as the user seeks a goal or reward. The agent uses information in the neural signal to create sets of movement commands (actions) for the prosthetic, while it learns the optimal neural state to action mapping for maximizing rewards.
This transormative framework allows users and computers to co-evlove with eachother through experience. See the paper for more details.
NSF Project: Dynamic data-driven brain-machine interfaces (PI: Fortes)
Description: The long term objective is to advance the understanding of motor-control brain models to enable the design of effective Brain-Machine Interfaces (BMIs) for humans with motor behavior limitations (e.g. paraplegics). A BMI uses a computer to implement brain models that translate signals from individual neurons into artificial limb commands. Discovery of the knowledge needed to uncover the unknown aspects of systems-based neural encoding and decoding for complex tasks needs highly demanding computational modeling. The architecture consists of multiple forward-inverse pairs of dynamic models for movement planning and control. The movement commands are the combined outputs of selected pairs of models on the basis of real-time feedback signals. The research aims to (1) identify the types, numbers and combinations of models for complex movement control and (2) deploy the cyberinfrastructures for both BMI implementation and research. It uses closed-loop experiments where a computer processes brain signals from rats to control robotic movements.
Instrumentation needs: Depending on the system scope, hundreds or even thousands of models (and processors) may be needed for brain modeling. One goal is to reduce these needs and identify system designs whose size, power consumption and cost are affordable by patients and researchers. Towards this goal, BMI research must have access to significant computing resources. It must be possible to aggregate and reserve the necessary resources, create parallel execution environments and guarantee the real-time response needed for low latency between brain signaling and sensory feedback. It must be possible to embed on-demand a BMI research test bed in a combination of shared and dedicated resources. The experimental test bed (below), is distributed across two laboratories in different buildings of the UF campus. It uses remotely located computers and a web interface to configure and monitor live experiments. Real-time computation and model adaptation takes less than 100 milliseconds. The resources are also used to run offline studies using data collected during live experiments.
The team includes: Jose Fortes, Justin C. Sanchez, Jose C. Principe, Renato Figueiredo, Linda Hermer-Vasquez (more info)
NIH Project: An Ultra-Low Power Wireless Neural Recording Implant Based on a Novel Pulse Representation (PI: Harris)
Description: The overall goal of this project is to design a new generation of fully implantable flexible substrate microelectrode array probes to record neural activity from behaving animals. A fully implantable unit would allow improved characterization of brain function via neural recordings in animals in an unrestrained condition. The proposed device is a battery powered electronic chip that utilizes the state-of-the-art integrate-and-fire (IF) representation and proven protocols: microwire array, flexible substrate, amplifier, and wireless communication. What makes this implantable specification possible is a new IF sampling principle that is able to reduce both the power dissipation and the necessary bandwidth to transmit high resolution data. We anticipate that it is possible to build an implant that uses less than 2 mW of total power dissipation to record, amplify, encode and transmit wirelessly raw neural recordings (field potentials and extracellular neuronal firing), which enables a battery powered, completely subcutaneous probe for behavioral experiments. An external signal reconstruction algorithm will output neural data with at least 40dB accuracy (better on high amplitude signal regions) at a 20 kHz sampling rate. We are in the process of designing, characterizing, building and testesting in vivo the recording system call the Florida Wireless Implantable Recording Electrodes (FWIRE).
The team includes: Harris, Sanchez, Nishida, Bashirullah, Principe
