We report an ongoing work on a biologically inspired device, the Neurosolver, introduced in our (1995) earlier paper. To explore and improve the Neurosolver's capabilities, we attempt to apply it to a task of rearranging three different blocks in a blocks world similar to the worlds known from the classic AI literature. The Neurosolver records the observed trajectories in the state space of the blocks world and uses the learned traces to perform searches and construct plans to control the movements of the blocks. This work is a part of a broader effort to devise neurally-based agents that would cooperate en masse in a process of solving complex problems. We consider it a next step toward a Neuromorphic General Problem Solver.
展开▼
