Using a hierarchical control architecture based on one conjectured to be present in living organisms, we have built a physical simulation of a walking robot with four or more legs, which exhibits very lifelike movement and purposive activity. It is able to:
- Stand up and balance on uneven terrain
- Resist random external forces
- Seek out virtual food particles
- Continue to operate (less robustly) when some legs are removed.
The robot achieves all this without requiring any of the following:
- A model of its environment
- A model of itself
- Motion planning
- Inverse kinematics
- Vision or other complex senses.
A simple physics engine simulates the robots body, its sensors, and its actuators. A variety of actuators have been simulated: idealised ones that supply a specified torque to a joint, and more realistic models such as pneumatic pistons. A hierarchical control architecture uses the actuators to control the robots posture, as sensed by the sensors. The robot senses only the height of its body from the ground, its joint angles and their rates of change, the positions of its feet relative to its body and with a pair of antennae, the strength of a scent signal at each antenna, which is radiated by each food particle. The robot is of interest from three points of view: Robotics, this design exhibits impressive performance relative to the state of the art in walking robots. Control theory, the robot is a case study in solving a complex, nonlinear, multi-dimensional control problem with a very simple control architecture. Animation, the simulation generates lifelike, purposive movement.