Source-linked AI summary
OpenSpace Lab Solution to the IROS 2026 Indoor Exploration Competition
Yuxuan Zhang, Dong Li, Zezhou Sun, Yuxuan Xu, Siyu Teng, Yuchen Li, Jianjian Yang, Long Chen
TL;DR
Time-bounded exploration must balance information gathering with the cost of returning to upload collected maps. OpenSpace Lab combines map-completion-based planning and homing-aware decisions for single robots with shared-map and intent-based coordination for teams. The solution ranked first in the public single-robot track, with 61.04% average coverage.
Problem
Time-bounded exploration requires robots to maximize information gathering while accounting for return and data-delivery costs, but explicit hard time constraints in frontier evaluation and planning remain largely unaddressed.
Source: §1 · p. 1Method
The approach uses pre-trained map completion for single-robot global planning, integrates homing constraints into remaining-time decisions, and coordinates multi-robot targets using shared map and intent information.
Source: §1 · pp. 1–2, §2.1.1 · p. 2, §2.1.3 · p. 3Results
61.04% average coverage rate ranked the solution 1st overall across the seven public maps in the preliminary stage.
Source: §3 · p. 5Takeaways & Limitations
Abstract
from arXiv · showhide
This report presents the \textbf{OpenSpace Lab}'s solution to the Competition on Intelligent Information Gathering for Single and Multi-Robot Systems Workshops, organized as part of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2026. Our team reached 1st place in the Single-Robot Public Track and 3rd place in both the Single- and Multi-Robot Private Tracks. The single-robot framework utilizes pre-trained map completion predictions for global planning to prioritize unexplored areas. To reconcile map coverage with limited operation time, we introduce a remaining-time-based exploration strategy that integrates homing constraints into the decision-making process. For multi-robot exploration, we utilize a utility-driven target selection strategy that balances observation gains, movement costs, and budget constraints, leveraging shared map and intent data to eliminate redundant search and maximize coordination efficiency. Our solution reached a 61.04\% coverage rate in the Single-Robot Public Track, while reaching 39.53\% and 39.91\% coverage in the Single- and Multi-Robot Private Tracks, respectively. An extended full-length paper based on this report is currently being prepared for submission, and the source code will be released upon acceptance of the full manuscript at https://github.com/OpenSpace-Lab/Indoor-Exploration-IROS2026.
1. Introduction
The introduction frames time-bounded exploration as a gap in which robots must gather information while accounting for the cost of returning to upload it. OpenSpace Lab presents separate single- and multi-robot strategies that incorporate map predictions, remaining-time decisions, and teammate information.
1. Introduction
- Existing exploration methods largely treat time or path length as post-mission metrics, while explicit hard time constraints in frontier evaluation and planning remain largely unaddressed.The introduction emphasizes that robots must account for the time needed to return to a base or communications hub and deliver collected data.Source: §1 · p. 1
- The competition imposes scale-dependent step budgets and requires robots to return within base-station range to upload map data.Source: §1 · p. 1
- The single-robot strategy uses pre-trained map completion to prioritize exploration sequences and dynamically adapts goal selection and pathing to varying step budgets.Source: §1 · pp. 1–2
- The multi-robot extension reduces redundant search through dispersed initial directions, shared teammate goals, and historical trajectories.Source: §1 · p. 2
2. Method
The method combines distinct single- and multi-robot pipelines with lightweight heuristics for exploration under strict step budgets. The single-robot pipeline uses map completion and global task ordering alongside local planning, while the multi-robot extension coordinates agents using shared information and homing estimates.
2. Method
- The two-tier architecture adapts separate single- and multi-robot pipelines to their operational dynamics, coupling algorithmic formulations with lightweight heuristics for strict step budgets.Source: §2 · p. 2
2.1. Single-Robot Method
- The single-robot framework combines learning-augmented global task ordering with local planning and adapts execution heuristics to different step budgets.Source: §2.1 · p. 2
2.1.1. Global Task Planning and Local Execution
- The global planner uses LaMa within MapEx for map completion, groups frontiers into macroregion tasks, and uses beam search to score visiting sequences under travel, exploration, homing, and transition-risk costs.Budget feasibility is checked during beam expansion; local planning validates the target against an observation-based baseline and issues immediate waypoints.Source: §2.1.1 · p. 2
2.1.2. Budget-Aware Engineering Heuristics
- Budget-aware heuristics vary by step limit, including route-overlap thresholds and delayed lateral-branch exploration, with a late-stage distance advantage for 2,000-step runs.The excerpt specifies 75% route overlap for short-budget runs, 16% for medium-budget runs, and a late-stage weight of 0.18 for long-budget runs.Source: §2.1.2 · pp. 2–3
2.1.3. Homing Condition and Data Upload
- Final homing begins when remaining steps fall below estimated return cost plus a 10-step safety margin; the estimate accounts for reaching communication range and the base station.Communication can be restored and the map uploaded once line of sight to the base station is re-established, before physically reaching the base grid.Source: §2.1.3 · p. 3
2.2. Multi-Robot Extension
- Multi-robot target selection balances ray-cast information gain and path costs against teammate intent and trajectory penalties, with initial angular dispersion and a homing-budget constraint.Agents also use event-triggered relays and dynamic handoffs; at critical budget limits or without a return path, they switch to final homing.Source: §2.2 · pp. 3–4
3. Experiments
The evaluation compared preliminary performance on public maps with a final benchmark on unreleased private maps. OpenSpace Lab ranked first in the public single-robot track and third in both private tracks.
3. Experiments
- Evaluation used public maps for preliminary tuning and unreleased private maps for a single-attempt final benchmark, with four starting locations per map scale.Single-robot budgets were 1,000, 1,500, and 2,000 steps for small, medium, and large maps; multi-robot teams of 2–5 robots were evaluated with 500, 1,000, and 1,500 steps.Source: §3 · pp. 4–5
- 61.04% average coverage rate earned 1st overall across seven public maps, with a 3.53-percentage-point lead over 2nd place on large maps (49.81% vs. 46.28%).Table 2 reports results for the single-agent public track.Source: Table 2 · p. 5, §3 · p. 5
- 3rd place in both private tracks included 1st place in the single-robot medium-scale category at 42.53% coverage.The final evaluation used ten unreleased private maps.Source: §3 · p. 5
4. Conclusion
The report presents OpenSpace Lab’s single- and multi-robot exploration strategies and competition results. A comprehensive algorithm-design manuscript is in preparation, and the codebase and enhanced implementations are slated for release upon paper acceptance.
4. Conclusion
- The report presents OpenSpace Lab’s single- and multi-robot exploration strategies alongside its competition results.Source: §4 · p. 5
- A comprehensive algorithm-design manuscript is being prepared for formal peer review.Source: §4 · p. 5
- The competition codebase and enhanced algorithm implementations will be open-sourced upon paper acceptance.Source: §4 · p. 5