Back to Rainer’s Home Page

 

 

How hierarchical control self-organizes in artificial adaptive systems

 

Rainer W. Paine, Jun Tani

RIKEN Brain Science Institute, Laboratory for Behavior and Dynamic Cognition, 2-1 Hirosawa, Wako-shi, Saitama, 351-0198, Japan

November, 2004

 

 

Abstract

 

Diverse, complex, and adaptive animal behaviors are achieved by organizing hierarchically structured controllers in motor systems. The levels of control progress from simple spinal reflexes and central pattern generators through to executive cognitive control in the frontal cortex.  Various types of hierarchical control structures have been introduced and shown to be effective in past artificial agent models, but few studies have shown how such structures can self-organize.  This study describes how such hierarchical control may evolve in a simple recurrent neural network model implemented in a mobile robot.  Topological constraints on information flow are found to improve system performance by decreasing interference between different parts of the network.  One part becomes responsible for generating lower behavior primitives while another part evolves top-down sequencing of the primitives for achieving global goals.  A hierarchical neural network is shown to outperform a comparable single-level network in controlling a mobile robot.

 

 

 

Figure 1  Maze environment. Shows trajectory for a Left-Right-Right sequence.  A simulated mobile robot  (Khepera II) learns to reach 8 different goals starting from a home position.  Two types of robot controllers (Figure 2) are evolved using a standard Genetic Algorithm.  The simulations are run using Webots 4 software.  Click HERE for a demonstration.

 

 

 

 

 

         

Figure 2  The two types of controllers evolved for the robot in Figure 1.  A standard fully-connected Continuous Time Recurrent Neural Network (CTRNN) with 11 neurons is shown in (B).  Inputs from the robot’s 8 infra-red sensors are received by 5 neurons.  Two neurons send motor commands to the two wheels of the Khepera robot.  This single network can encode multiple movement sequences.  The initial activity states of the two “task” neurons determine which three-turn movement sequence the robot will execute.  In the so-called “bottleneck” network shown in (A), the neural network has been subdivided into top and bottom levels, such that information flow between the two levels is limited by the two “bottleneck neurons”.  It was found that the bottleneck network was able to find more goals on average than the standard fully-connected network.  Further, different functions were automatically adopted by the two levels of the bottleneck network.  The bottom level took over control of obstacle avoidance (so the robot wouldn’t hit any walls) as well as left/right turning behavior.  The top level took over control of sequencing the turns in order to find particular goals.  Since obstacle avoidance requires the robot to respond quickly to changing sensory inputs, the bottom level neurons evolved to change rapidly to incoming signals, as shown by the motor output neurons in the bottom of Figure 3.  Further, since the overall sequence task took a much longer time than short-term obstacle avoidance, the task neurons of the top level evolved to change slowly in response to incoming signals (Figure 3, top). 

Figure 3  Neuronal activity for a Right-Left-Right turn sequence in the bottleneck network.  Top: Neuronal activity of bottleneck and task neurons with slower evolved time constants to encode turns and sequences, respectively; Bottom: Activities of motor output nodes that evolved faster neuronal time constants to respond to immediate task demands.

 

 

 

 

 

Figure 4  Phase space analysis for bottleneck neuron activities.  They determine turn direction at intersections of the maze in Figure 1.  When the activities of the bottleneck neurons are in the black area, the robot turns left at a corner.  When they are in the white area, the robot turns right.  In the grey area, the robot collides with a wall because the turning behavior encoded by the bottleneck neurons becomes too strong and overrides the obstacle avoidance behavior encoded in the bottom level of the network. 

 

 

 

 

 

 

 

 

 

 

A                                                         

 

 

 

 

 

 

 

B

Figure 5  Phase space analysis of task neuron initial activity states and resulting turn sequences.  For example, when the task neuron initial states are in the orange region at the beginning of movement, then the robot will make a Right-Left-Right turn sequence, leading it to goal number 6 in Figure 1.  A single neural network can thus encode multiple movement sequences.  The initial states of two neurons, the so-called “task neurons” in Figure 2, determine which movement sequence the robot will make.  The turn decision tree structure (A) corresponds to (B) in which six goals of the maze in Figure 1 were found.  Letters represent turns of the sequences (L=Left, R=Right).

 

Figure 6  Left-Right-Left trajectory of a real Khepera II robot.  The robot controller uses the bottleneck architecture of Figure 2.  The controller was first evolved in simulation and then transferred to the real robot.

 

Conclusion

     This work has demonstrated an approach to encoding goal-directed, behavioral sequences in a self-organized recurrent neural network controlling simulated and real mobile robots.  It further examined how hierarchical segregation of control can emerge in a given architecture and enhance controller performance.  Different types of dynamic structures self-organize in different levels of the network for the purpose of achieving complex navigation tasks.  Top-down behavioral switching emerges through parametric bifurcation of lower level activity via bottleneck neurons.  In the higher level, a topologically ordered mapping of initial cell activity states to motor-primitive sequences self-organizes by utilizing the initial sensitivity characteristics of nonlinear dynamic systems.  Task-neuron modulation could be effected by even higher level networks which could represent sequences of sequences for different sets of ever more complex tasks.  This research serves as an example of how complex dynamic structures with initial sensitivity and task-dependent temporal activity may self-organize to control simpler structures that encode movement primitives.  Such structures may be analogous to those which encode movement sequences in biological neural networks, and may be a promising direction for research into mobile robot navigation. 

 

For further information, please see my Publications.

 

Back to Rainer’s Home Page