Brain Based Devices - Artificial bodies with artificial nervous systems
NSI home Job Opportunity Brain-Based Devices Publications

Brain-Based Devices

Over the past sixteen years, researchers at The Neurosciences Institute in La Jolla, California, have designed a series of theoretical models, known as Brain-Based Devices (BBDs), to investigate nervous system function. A BBD is a realistic brain model that controls a robotic device performing a behavioral task.

The Institute's researchers believe strongly that the brain does not function in isolation. An organism's brain is closely coupled to its body which actively interacts with its environment. Using BBDs we can perform tests with all of the noise and sophistication of the real world and record the activity from its entire brain, something that is not yet possible to do with live animals.

History and Accomplishments

DIII image
Darwin III learned to control a virtual arm in order to knock away "noxious" objects.

The Institute's work has yielded revealing insights, e.g. into the importance of self-generated movement in the aquisition of perceptual capability and the role of value or reward systems in adaptation and learning. We have also been able to show that the unique anatomy of a brain region, such as hippocampus, is important for its observed neural function.

The Darwin series of neural automata began in 1981, and were originally software models. Since 1992, these BBDs have had physical bodies that interact with the environment.

DXI image
Darwin XI learned to navigate a maze and developed responses in its simulated hippocampus that are similar to those observed in rats.

In 2000 we developed a BBD body known as NOMAD, for Neurally Organized Mobile Adaptive Device. The NOMAD platform has many sensors, such as a pan-tilt color camera for vision, artificial whiskers for texture sensing, a compass and wheel-encoders for a sense of head direction and self-movement, infra-red transceivers and a laser rangefinder which provide a sense of proximity to objects. NOMAD moves autonomously about its environment in real-time, its behavior controlled by a simulated nervous system running on a set of powerful computers (a Beowulf cluster of up to 64 CPUs). These simulated nervous systems have realistic neuroanatomy, on the order of a hundred thousand neuronal units, and a few million synaptic connections between those units.

In 2004 work began on the Segway platform, a BBD body based on the commercially-available scooter, which is capable of operating outside of a controlled laboratory environment. We are currently working on this and other BBDs that operated in unconstrained, real-world environments.

  • Darwin X and Darwin XI (2005-2007) investigated a hippocampal model of spatial, episodic, and associative memory that learned to navigate both open-field and maze environments. More information on Darwin X-XI
  • Segway Cerebellum Model (2006) This model of the cerebellum was constructed on a Segway platform and investigated predictive motor control in an obstacle avoidance task. More information on the Segway Cerebellum Model
  • Segway BBD
    The simulated nervous system of the Segway Soccer Playing BBD enabled it to safely and skillfully play a soccer-like game, interacting with both human and other robotic players

  • Segway Soccer Playing BBD (2005) played a soccer-like game where teams consisted of both human players and other robots. This model was a hybrid system, where a simulated nervous system was used in conjunction with more traditional robot control techniques to produce skilled, reliable, and safe behavior. More information on the Segway Soccer Playing BBD
  • Darwin IX (2004) had artificial whiskers and to model texture discrimination in somatorsensory cortex.
  • Darwin VIII (2004) modelled how reentrant connections in visual pathways and neural synchrony can bind features of complex, overlapping objects into a unified scene.
  • Darwin VI and Darwin VII (2000-2002) were the first BBDs to split up the sense-perceive-act loop into concurrent, real-time processes. These models explored the role of history in perceptual categorization, and demonstrated sophisticated invariant visual object recognition. More information on Darwin VI-VII
  • Darwin V (1998) demonstrated that translation invariance and pattern selectivity emerge due to the continuity of self-generated movement.
  • Darwin IV (1992) performed a tracking and conditioning task and applied sythetic neural modeling to a real-world device for the first time.
  • Darwin III (1990) was the first synthetic neural model of an organism interacting with an environment while engaged in a sensorimotor coordination task. The simulated nervous system with vision and tactile sensing learned to control a virtual arm in order to knock away "noxious" objects.
  • Darwin II (1982) demonstrated the importance of reentrant connections in neuronal group selection during a categorization task.
  • Darwin I (1981) demonstrated the principles of degeneracy, amplification, and selection in a pattern recognition task.

More Information:
Darwin X/XI
A model of episodic, spatial, and multimodal memory formation
Segway Cerebellum
A model of predictive learning
Segway Soccer Playing BBD
Interacting with humans on the field
Darwin VII
A BBD performing a classical conditioning task