Adaptive Neural Oscillator with Synaptic Plasticity Enabling Adaptive Hexapod Locomotion
Georg-August-Universität Göttingen, Drittes Physikalisches Institut, Germany
There is strong evidence and it is widely accepted that the rhythmic neural activity necessary for animal locomotion is basically produced by central pattern generators (CPGs). They can generate periodic patterns without requiring periodic input or sensory feedback. However, sensory feedback still plays an important role when it comes to coordination of multiple joints or limps and adaptation to changing environments.
Recently, we developed an adaptive neural oscillator with synaptic plasticity that is able to adapt to and memorize an external frequency quite fast and works within a wide frequency range . It is based on a discrete-time two neuron SO(2)- network and inspired by a general oscillator adaptation rule. A third neuron with additional plastic synapses is introduced to accelerate the adaptation process.
We use this mechanism as a CPG to control the joints of a simulated hexapod robot. The sine shaped output of the CPG is fed into a modular feed-forward post-processing network that generates appropriate motor commands to control the limbs of the robot with the given frequency. Due to the limited torque of the motor a delay between the angle set point and the actual motor position is observed. The signals of the forward-backward joints' angle sensors are used as a feedback signal to the adaptive CPG. By trying to make the feedback signal and the CPG output coincident, the adaptation mechanism maintains a given phase delay between the motor command and the actual angle sensor value.
The phase delay between the motor command and the angle sensor signal depends on the motor power, the current workload and the frequency at which the motor is driven. For higher frequencies, lower motor power or higher workload the phase delay increases. As a consequence the CPG converts to a lower frequency. The same holds for the opposite case. If the robot reaches e. g. an upwards slope, the workload increases and so does the phase delay. Therefore, the CPG frequency decreases until the original phase delay is recovered. If the robot reaches a negative slope the opposite effect can be observed. This behavior is reasonable as only a lower frequency enables the robot to climb upwards efficiently. In contrast, a higher walking frequency can be used when walking downwards without wasting energy.
As a further finding we observe that blocking the robot's movement, e. g. by applying a large downwards force, effectively results in a huge phase delay as the motors nearly cannot follow motor commands at all. This leads to a reduction of the oscillation frequency towards zero which is a useful behavior to prevent motor damage.
The application of the recently developed neural oscillator with synaptic plasticity as a CPG for a simulated hexapod robot results in autonomous and meaningful adaptation of the robot's walking speed when confronted with terrains with different slopes.
This research was supported by Emmy Noether grant MA4464/3-1 of the Deutsche Forschungsgemeinschaft and Bernstein Center for Computational Neuroscience II Göttingen (BCCN grant 01GQ1005A, project D1).
 Nachstedt, T., Wörgötter, F, and Manoonpong, P. (2012). Adaptive Neural Oscillator with Synaptic Plasticity Enabling Fast Resonance Tuning. Submitted.
central pattern generator,
recurrent neural networks,
short-term synaptic plasticity
Bernstein Conference 2012, Munich, Germany, 12 Sep - 14 Sep, 2012.
Motor control, movement, navigation
(2012). Adaptive Neural Oscillator with Synaptic Plasticity Enabling Adaptive Hexapod Locomotion.
Front. Comput. Neurosci.
Bernstein Conference 2012.
11 May 2012;
12 Sep 2012.
Mr. Timo Nachstedt, Georg-August-Universität Göttingen, Drittes Physikalisches Institut, Göttingen, 37077, Germany, email@example.com