Neural Systems Lab
Our research focuses on understanding the brain using computational models and simulations, and applying this knowledge to the task of building intelligent robotic systems and brain-computer interfaces (BCIs). We utilize data and techniques from a variety of fields, ranging from neuroscience and psychology to machine learning and statistics. Current efforts are directed at: (1) understanding probabilistic information processing and learning in the brain, (2) building biologically-inspired robots that can learn through experience and imitation, and (3) developing interfaces for controlling computers and robots using brain- and muscle-related signals.
News/Announcements
-
December, 2006: We have developed a non-invasive brain-computer interface (BCI) that allows a user to command a humanoid robot to pick up objects and bring it to specific locations. A humanoid robot can use sophisticated robotics and computer vision techniques to explore the environment, discover possible objects and interactions, and interact with the objects selected by a user. In this way, a simple BCI-based selection interface can enable a user to perform significantly complex actions.
- Press Coverage:
- Link to more details.
- Video of the BCI-controlled Robot, and in Quicktime format.
- Video describing the brain processes exploited by the BCI interface.
-
July 12, 2006: We have developed a new method for robotic imitation of human motion based purely on learning from experience and observation. The method uses a sensory-motor mapping in a low dimensional space to reduce computational complexity, allowing optimization of postures during motion to match robot dynamics. Our results demonstrate that a humanoid robot can learn to walk by observing a human. The first row of the figure below shows a human subject demonstrating a walking gait through a motion capture system. The second row shows simulation results of the robot's motion without learning. The third row shows simulation results after learning and the last row shows the results on the real robot. Further details of this research can be found here. Related publications:
-
July 30, 2005: We have recently been studying robot imitation and closely looking at methods to map between the high-dimensionality space of robot movements to the equally high-dimensionality space of human movements.
The figure below shows corresponding robot/human poses from various actions and the intervening 2d latent space.
- April 30, 2005:
- Article on EMG-based mappings for robotic arm control, to appear in Proceedings of AAAI, 2005.
- Experimental session video Real-time classification of EMG into 8 classes for continuous control of a robotic arm.
- Sorting Lego Toys! Using the real-time EMG control scheme to reach for, pick up, and sort lego toys into separate piles.
- Photos of setup
- Videos and Images:
- January 5, 2005:
- An upcoming UWTV Broadcast on the Neural Systems Laboratory is now available. Further information can be found at www.uwtv.org.
- December 15, 2004:
- BCI/ECoG: Further information on our current
BCI experiments, including information on recent results. Click
here
to view the Seattle PI article on these experiments.
- HOAP2 soccer demo: A video demonstration of the HOAP2 tracking a soccer-ball and kicking it.
- BCI/ECoG: Further information on our current
BCI experiments, including information on recent results. Click
here
to view the Seattle PI article on these experiments.
-
October 5, 2004 : New Videos of the HOAP2
:
The following videos demonstrate some of the basic abilities of the HOAP2 to perform various tasks.- HeadTracking: Shows the use of a Viola-Jones detector to find faces in an input image and have the robot track the person.
- JoystickHoap: Shows the use of a joystick based controller to control individual joints in the HOAP2.
- ToolboxAdventure: A hybrid motion demo based on recorded movements from the joystick control program and a simple walking program that demonstrates some of the basic abilities of the HOAP2 including object manipulation tasks. A longer version is available here.
- WaveBye: A robotic farewell.
-
Videos of Robotic Gaze Imitation
- Final Test: Before each individual test the Biclops scans the table, to show the objects involved in each part of the experiment. During this series of tests, the subject (Aaron) looks at different objects laid on the table, and the Biclops is able to learn the saliency distribution online. Later in the series the objects are replaced with completely different objects, some with the same size and color, and the Biclops is then be able to look at and identify the most salient object, regardless of whether it has seen that specific object before.
Projects and Publications
- Probabilistic models of the brain
- Hierarchical Bayesian inference in networks of spiking neurons (Advances in NIPS 17, pp. 1113-20, 2005).
- Bayesian computation in recurrent neural circuits (Neural Computation, 16(1), 1-38, 2004)
- Probabilistic Models of the Brain: Perception and Neural Function
- Other publications
- Probabilistic imitation learning
- Robotic Imitation from Human Motion Capture using Gaussian Processes
- A Bayesian Model of Imitation in Infants and Robots (in Imitation and Social Learning in Robots, Humans, and Animals: Behavioural, Social and Communicative Dimensions, K. Dautenhahn and C. Nehaniv (eds.), Cambridge Univ. Press, 2004)
- A Probabilistic Framework for Model-Based Imitation Learning (CogSci 2004) (slides available here)
- Spike-based computing and learning
- Learning algorithms for vision
- Brain-computer interfaces
- Dynamic Bayesian networks for brain-computer interfaces (Advances in NIPS 17, pp. 1265-72, 2005)
- EEG-based systems for communication and control
- ECoG-based systems for epilepsy patients
- EMG-based Mappings for Robot Arm Control
Collaboration
- Andy Meltzoff, Institute for Learning and Brain Sciences, UW
- Jeff Ojemann, UW Medical School
- Michael Shadlen, Physiology and Biophysics, UW
Laboratories
- Humanoid Robotics Lab, Allen Center 286
- Brain Computer Interface Lab, Allen Center 208
Funding Provided by:
Office of Naval Research |
National Science Foundation |
also:
Alfred P. Sloan FoundationDavid and Lucile Packard Foundation