| 摘要 |
A neural controller, to be called a neuroplanner, is described herein for finding a sequence of system configurations from an initial system configuration to any one of a set of goal system configurations specified by a targeting system the presence of prohibited system configurations. Means are provided for the goal system configurations to be specified by selecting a targeting configuration that is in turn correlated with one or more system configurations. In an embodiment as a robot arm controller, the system configurations are arm configurations, the targeting system is a pointing device that specifies a point in three dimensional space that is to be touched by the endeffector, and the prohibited system configurations are illegal arm configurations; e.g., those that place the arm beyond the physical limits of a joint of the robot arm or in a region that must be temporarily avoided because it is currently occupied by an obstacle. The sequence of configurations provided by the controller constitutes a trajectory that will move the arm from its current configuration to any configuration in which the end-effector is at the specified target point while simultaneously avoiding obstacles. The neuroplanner includes means for creating an environment network, a system configuration network, and a targeting configuration network that respectively encode quantized representations of the environmental information, the system configurations, and the targeting configurations. Furthermore, means are provided for correlating the environmental network with the system network and also correlating the targeting network with the system network so as to be able to specify prohibited system configurations and goal system configurations respectively. Finally, means are provided for finding a path from one system configuration to a goal system configuration that excludes prohibited system configurations.
|
