On-line assembly planning for stochastically reconfigurable systems

Abstract

Stochastic assembly approaches can reduce the power, computation, and/or actuation demands on assembly systems by taking advantage of probabilistic processes. At the same time, however, they relinquish the efficiency and predictability of deterministic alternatives. This makes planning error-free assembly sequences challenging, particularly in the face of changing environmental conditions or goals. Here we address these challenges with an on-line approach to assembly planning for stochastically reconfigurable systems where the spatial and temporal availability of modules is uncertain, either due to a stochastic assembly mechanism, resource fluctuations, or large numbers of uncoordinated agents. We propose an assembly algorithm that is guaranteed to find an assembly path for finite-sized, connected objects. This is achieved by sampling the space of possible assembly paths to the target structure that satisfy assembly constraints. Assembly is accelerated by pursuing multiple paths in parallel. The algorithm computes these parallel assembly paths on-line during assembly and is thus able to adapt to changing conditions, as well as predict the remaining assembly time. For situations where the number of paths found exceeds the number that can be pursued in parallel, the assembly algorithm further maximizes assembly rates according to domain-specific local assembly costs.