BIC-GA: Return-Aware Cooperative Exploration in Fragmented Environments via Directed Commitments and Gate Admission
Abstract Dual-robot exploration in fragmented unknown environments is constrained by incomplete sensing and complex free-space structure. In narrow passages, dead ends, and shared bottlenecks, frontier updates can destabilize assignments, while locally attractive targets may be infeasible once branch entry, service execution, and return requirements are considered. This study presents Branch-Intent Commitment with Gate Admission (BIC-GA) for return-feasible dual-robot target-coverage exploration. Instead of assigning raw frontier points, BIC-GA models each frontier component as a structured, directed commitment characterized by its entrance location, traversable entry orientation, expected information gain, entry difficulty, and predicted service workload. The robots select a commitment pair through a return-aware joint objective after predictive loop-budget screening, considering outbound, service, and return costs, overlap, short-prefix conflict, and workload balance. At local gates, exploratory changes require sufficient improvement over legal continuation, while an execution guard checks post-step return margins before motion to preserve modeled return feasibility. Experiments on fragmented grid maps show that, over jointly successful conditions, BIC-GA reduces mean team path length by 6.2–26.4%, turning effort by 34.9–50.4%, and composite motion cost by 7.8–27.8% relative to four tested baselines. Component tests further demonstrate more stable assignments, fewer reversals and conflict-repair interventions, and reliable exploration–return performance under budget constraints.
Authors
- Chee Keong Tan (ORCID: https://orcid.org/0000-0001-5551-1017)
- Yao Xue (ORCID: https://orcid.org/0009-0002-0460-3880)
Institutions
- Monash University Malaysia (MY)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-13
- DOI
- https://doi.org/10.21203/rs.3.rs-11248079/v1
- Primary Topic
- Robotic Path Planning Algorithms
- Type
- preprint