Modular robot programming can be substantially more challenging than normal robot programming due to:
• Scale/number of modules.
• Concurrency and asynchronicity, both in physical interactions and potentially at the software level.
• The local scope of information naturally available at each module.
Recent declaration approaches such as P2 and Saphira have shown promise in other domains that share some of theses characteristics. Inspired by those results it has been developing two modular robot specific declarative programming languages, Meld and LDP. Both languages provide the illusion of executing a single program across an ensemble, while the runtime system of each language automatically distributes to computation and applies a variety of optimizations to reduce the net computational and messaging overhead.
It has previously described a whole motion based shape planning algorithm that exhibits constant planning complexity per module and requires information linear in the complexity of the target shape. Subsequently it generalized this approach to extend its functioning to other local metamorphic systems. The latter, generalized algorithm operates on sub-ensembles to accomplish both shape control and resource allocation while maintaining global connectivity of the ensemble.
Meld has proven more effective at many of the global coordination aspect of this algorithm, at efficiently tracking persistent operating conditions, and at coping with non linear topologies. LDP has proven more effective at local coordination, sophisticated temporal conditions, detecting local topological configurations, and more generally, at expressing variable-length conditional expressions.
No comments:
Post a Comment