In plain words
Robots moving together need plans that account for one another. A planner that plans for the whole team can work offline, while robots planning separately in real-time may settle for a less effective team solution.
WarpWeaver has each robot search possible motions, then share promising command sequences and account for space other robots need. In 11 robot motion-planning tasks, its median team travel time was lowest on 10 compared with centralized offline planners.
The research
Can a robot team seek a strong joint plan while each robot plans locally and the team must act in real time? Centralized planners can pursue joint optimality but run offline; decentralized real-time planners often settle for local optima.
WarpWeaver has robots search locally and negotiate shared-space costs, aiming for near-optimal team plans in real time.
Can robots plan separately in real-time yet approach a good team-wide solution? WarpWeaver has each robot search possible motions and exchange promising command sequences, using shared-space prices to account for conflicts with peers.
On 11 Dynobench robot motion-planning instances, WarpWeaver completed every trial and had the lowest median total travel time summed across robots on 10 instances, compared with centralized offline planners.
The result supports the approach on those tasks; the largest reported reduction was 22%, and the planners differed in both design and real-time versus offline execution.
The formal near-optimality certificate applies only when the negotiation reaches quiescence and under the paper’s smoothness and constraint-margin assumptions.
The experiments cover specific simulated tasks and six-robot demonstrations; the authors say tightly coupled tasks such as rigid-formation maintenance may be harder to solve.
Analysis
Deep
- Original paper authors
- Shashank A. Deshpande, Blake T. Edwards, Jonathan P. How
- Submitted to arXiv
- License
- arXiv non-exclusive