Minimalism Distribution Supermodularity

We have designed and implemented multi agent strategies for manipulation tasks by distributing mechanically based sequential algorithms across sev eral autonomous spatially separated agents such as mobile robots Our experience using mobile robots for the manipulation of large objects couches boxes le cabinets etc leads us to recommend a minimalist architecture for multi agent programming In particu lar our methodology has led us to derive asynchronous distributed strategies that require no direct communi cation between agents and very sparse geometric and dynamic models of the objects our robots manipulate We argue for a design principle called supermodu larity which is orthogonal both to the notion of modu larity in cognitive AI and also to horizontal decompo sition the non modularity advocated in the subsump tion connectionist literature Finally we discuss a simple mobot Scheme in frastructure to implement supermodular architectures In the past few years we have programmed many su permodular manipulation protocols and tested them ex tensively on our team of mobile robots We describe why we think the supermodular infrastructure results in robust simple readable manipulation strategies that can be recycled and reused