Event Abstract Back to Event A Distributed Cognitive Map for Spatial Navigation Based on Graphically Organized Place Agents Jorg Conradt1* and Rodney Douglas1 1 ETH & University Zurich, Switzerland Animals quickly acquire spatial knowledge and are thereby able to navigate over very large regions. These natural methods dramatically outperform current algorithms for robotic navigation, which are either limited to small regions [1] or require huge computational resources to maintain a globally consistent spatial map [2]. We have now developed a novel system for mobile robotic navigation that like its biological counterpart decomposes explored space into a distributed graphical network of behaviorally significant places. Each such a place is represented by an independent “place agent” (PA) that actively maintains the spatial and behavioral knowledge relevant for navigation in that place. The collection of such place agents does not represent space in a common consistent data structure as current topological or metric-based approaches do (Figure 1 left and middle). Instead, our system incrementally builds a graphical network consisting of PAs that each only knows its local space and its local nearest neighbors. Each of these PAs is unaware of its position within the network and the position it represents in global space. The topology of the network - which reflects the structure of traversable external space - only exists implicitly, as PAs only communicate with their direct local neighbors (Figure 1, right). No process in our system maintains or operates globally on the network; thus, it is only necessary to maintain spatial consistency locally within the graph. This simple strategy significantly reduces computational complexity, is robust to local perturbations, scales well with the size of the navigable region, and permits a robot to autonomously explore, learn, and navigate large unknown environments in real time. The system has explored a floor in our institute of 60x23m using a mobile robot, and created about 150 independent PAs to represent behaviorally relevant spaces. The resulting distributed network can demonstrate globally consistent navigation behavior without a global supervisor involved, e.g. guiding the robot to previously encountered stimuli. We expect that we can represent significantly larger environments without suffering degraded performance, as all PAs are independent programs that we can distribute among multiple computing units. Neural hardware is well suited to implement such a “distributed cognitive map”, as brains are intrinsically distributed processing systems without global access to all memorized information - which is required by traditional algorithms for navigation. We therefore are convinced that the proposed system resembles biological information processing much more closely than current methods for spatial navigation.