We propose a decentralized solution to a highlevel task-planning problem for a multi-agent system under a set of possibly dependent LTL f specifications. We propose an approach where the problem is turned into a number of individual two and a half player stochastic games with reachability objectives. If almost-surely winning strategies cannot be found for them, we deploy so-called least-limiting advisers to restrict agents' behaviours. A key step is treating safety and liveness separately, by synthesizing necessary safety and fairness assumptions and iteratively exchanging them in the form of advisers between the agents. We avoid the state-space explosion problem by computing advisers locally in each game, independently of the model and specification of other agents. The solution is sound, but conservative. We demonstrate its scalability in a series of simulated scenarios involving cleaning of an office-like environment.

We propose a compositional solution to the strategy synthesis problem for LTL specifications in the cooperative (heterogeneous) multi-agent scenario. A main challenge of the general strategy synthesis approach is the state-space explosion occurring during construction of a global model for agents with different, mutually dependent goals. Given a set of agents and their individual goal specifications represented through a local model and an LTL formula, we compute a compliant set of strategies that fulfill each agents' goal specification. We avoid the state-space explosion by computing individual solutions for each agent separately and then composing these solutions. During the initial strategy computation, assumptions over the states of other agents not represented in the local model are generated only where needed. These assumptions are resolved during composition of the individual solutions to assure compliance of the computed strategies. The effectiveness of this approach is demonstrated in several simulation case studies and compared to the classical, monolithic approach.

No abstract

Many tasks require robots to manipulate objects while satisfying a complex interplay of spatial and temporal constraints. For instance, a table setting robot first needs to place a mug and then fill it with coffee, while satisfying spatial relations such as forks need to placed left of plates. Existing solutions are often not robot-agnostic, not expressive enough to monitor and plan complex tasks, and require users to manually define object relations. We propose the spatio-temporal framework SpaTiaL that unifies the specification, monitoring, and planning of object-centric robotic tasks. SpaTiaL captures diverse spatial relations between objects and temporal task patterns. Our experiments with recorded data, simulations, and robots demonstrate how SpaTiaL provides real-time monitoring and facilitates online planning. SpaTiaL is open source and easily expandable to new object relations and robotic applications.