Source-linked AI summary
Obstacle-Aware Autonomous Coverage and Navigation for Outdoor Robots
Leonardo Gargani, Matteo Frosi, Matteo Matteucci
TL;DR
Long-duration outdoor coverage remains difficult because localization drift, obstacles, controller feasibility, and energy management challenge reliable field operation. The paper addresses this with a unified ROS 2 architecture combining dual-antenna RTK-GNSS localization, controller-aware coverage planning, and Nav2 Behavior-Tree execution. Across five outdoor areas, the system completed every route while sweeping 93.1% to 96.1% of the planned coverage area.
Problem
Reliable long-duration outdoor coverage requires more than geometric planning because localization drift, obstacles, turn-heavy controller constraints, and mission-level energy management remain challenging.
Method
A unified ROS 2 stack fuses dual-antenna RTK-GNSS and wheel odometry in an EKF, refines Fields2Cover for controller feasibility, and uses a Nav2 Behavior-Tree executive for persistent execution.
Results
93.1% to 96.1% of the planned coverage area was swept across five outdoor areas, with every coverage route completed.
Takeaways & Limitations
The integrated stack handled varied geometries, obstacles, recovery, and autonomous return-to-charge during the evaluated outdoor missions.
Takeaways & Limitations
The evaluation considers a single differential-drive robot on predominantly planar outdoor terrain, with known obstacles mapped before planning and runtime obstacles handled reactively.
Abstract
from arXiv · showhide
Long-duration outdoor coverage with autonomous platforms remains challenging beyond classical planning: deployments face localization drift in open spaces, obstacles in cluttered sites, controller feasibility in turn-heavy maneuvers, and persistent autonomy with energy management. We propose a unified ROS 2 architecture for outdoor coverage that combines coverage planning, robust localization, and Nav2-based execution. A dual-antenna RTK-GNSS fused in an EKF keeps the robot pose, both position and heading, accurate across long missions; three controller-aware refinements are added to a mature coverage planner; a Behavior-Tree mission manager coordinates multi-goal execution, layered recovery, cost-aware goal management, and autonomous docking for return-to-charge. We validate the stack through simulation and real-world trials across multiple outdoor areas with varying geometries and obstacle densities. Overall, these results show that the proposed stack can reliably complete outdoor coverage missions across varied areas, sweeping 93.1% to 96.1% of the planned coverage area.
1 Introduction
The paper identifies a systems-integration gap between geometrically valid coverage plans and reliable long-duration outdoor execution. It proposes a unified ROS 2 stack combining robust localization, controller-aware planning refinements, and Behavior-Tree mission supervision.
- Outdoor deployments require stable localization, obstacle handling, long-duration execution, and energy-aware return-to-charge beyond classical coverage planning.
- Turn-heavy, low-speed motion complicates heading estimation, while controller-infeasible waypoint distributions can make geometrically optimal paths fragile during execution.
- The proposed architecture integrates localization, perception, planning, and execution under a mission-level supervisor and is validated in simulation and five real outdoor areas.
- The localization stack fuses dual-antenna RTK-GNSS and wheel odometry in a custom 12-state EKF with sensor gating and direction-of-travel heading fallback.
- Three Fields2Cover refinements adapt waypoint density, preserve clearance during inter-swath transitions, and cluster cluttered obstacles into planning polygons.
- A Behavior-Tree executive supports batched multi-goal execution, layered recovery, cost-aware goal management, and marker-guided autonomous docking with energy-triggered return-to-charge.
2 Related Works
Related work covers geometric coverage planning, robust transitions and replanning, outdoor localization, and Behavior-Tree-based persistent autonomy. The paper combines these previously separate lines into one outdoor-coverage stack.
- Coverage path planning and field pipelines: Classical coverage methods use geometric decomposition, while practical planners commonly cover decomposed cells with parallel swaths.
- Coverage transitions and execution robustness: Prior work addresses inter-swath robustness through continuous-curvature turns, constrained headland optimization, anytime replanning, and online multi-robot planning.
- Outdoor localization and heading estimation: Outdoor coverage makes heading difficult because slow motion and frequent turns cause wheel-odometry drift over long missions.
- Outdoor localization and heading estimation: Dual-antenna RTK-GNSS provides motion-independent yaw that can be fused in an extended Kalman filter, with gating and travel-direction fallback for degraded measurements.
- Execution, behavior trees, and persistent autonomy: Nav2 supplies Behavior-Tree orchestration, layered costmaps, controllers, and recovery behaviors, while prior persistent-autonomy work includes charging and recovery.
- Execution, behavior trees, and persistent autonomy: The proposed stack combines these lines by deferring blocked goals to reactive navigation, adding localization safeguards, and tailoring planning refinements to a differential-drive platform.
3 System Description and Architecture
The paper defines an offline-to-online outdoor coverage architecture that converts site geometry into an admissible region and route, then executes it with localization, navigation, and Behavior-Tree supervision. Its formulation evaluates how much planned coverage is realized during execution, while the localization stack targets reliable pose estimation during long, turn-heavy missions.
- Problem formulation: The system models a planar differential-drive robot, known polygonal obstacles, runtime unknown obstacles, and a safety-expanded admissible region for collision-free planning.Known obstacles are available during planning; runtime obstacles are handled reactively. Obstacle growth enforces a margin δ ≥ ρ around the robot footprint.
- Problem formulation: Coverage plans are arc-length-parameterized pose curves whose width-w working band defines the ground region swept by the tool.The curve includes position and heading, and the swept region is determined by the implement width and robot orientation.
- Problem formulation: Coverage efficiency ηc measures the fraction of planned coverage area actually realized by the executed trajectory within the admissible region.Using the same implement width for planned and executed sweeps isolates execution quality from the plan; practical gaps to ηc = 1 reflect turning, overlap, and discrete waypoint following.
- Architecture and platform: Offline preparation produces the admissible region and coverage route, while online execution runs that route under a Behavior-Tree supervisor with reactive handling of unknown obstacles.The packaged stack supports reuse of offline-prepared routes across repeated runs of the same site.
- Architecture and platform: The onboard platform combines wheel encoders, dual-antenna RTK-GNSS, LiDAR, and a docking camera with a grid planner, smoothing, and regulated pure pursuit.The navigation stack uses a 1.0 m working tool, 0.9 m swath spacing, and a 20 Hz 12-state EKF.
- Localization: A custom 12-state EKF fuses wheel odometry with georeferenced RTK-GNSS position and motion-independent yaw, using gated measurements and fallback heading estimation.GNSS yaw is accepted above an orientation-confidence threshold; otherwise, heading can be inferred from consistent recent GNSS position increments, with filter prediction retained when directions are inconsistent.
4 Coverage Planning
The offline stage prepares a metric admissible region and plans a reusable Fields2Cover route, then refines waypoint sampling, inter-swath connectors, and obstacle geometry for execution.
- Admissible-region preparation: The offline pipeline prepares a metric admissible region from the site boundary and buffered known obstacles before coverage planning.Boundary acquisition yields a valid polygon, while obstacle footprints are transformed, buffered by δ, and subtracted from the boundary.
- Coverage route generation: Fields2Cover generates ordered parallel swaths over the admissible region using the working width and vehicle parameters.The planner reserves a headland, fills the interior with parallel swaths, and orders them to reduce non-productive travel.
- Collinearity-aware resampling: Controller-aware resampling preserves dense turn sampling while making straight swaths uniformly sampled without changing the traced path.Distance, decimation, and interpolation passes enforce spacing bounds while removing near-collinear vertices.
- Obstacle-aware connectors: Native looping connectors are replaced by straight goal-to-goal links whose actual trajectories are resolved online against the live costmap.This retains deterministic inter-swath structure offline while allowing the reactive layer to enforce clearance around known and unknown obstacles.
- Obstacle clustering: Nearby obstacles are merged morphologically to reduce fragile transitions through narrow gaps before planning.The obstacle layer is rasterized and dilated with a directionally sized rectangular mask tied to robot clearance needs.
5 Online Execution
The online stack uses a two-level Behavior Tree to execute routes, recover from navigation failures, defer blocked goals, and integrate energy-aware docking.
- Behavior-Tree mission executive: A two-level Behavior Tree coordinates mission execution from goal extraction and undocking through navigation and docking or low-battery return.The mission tree supervises the deployment loop, while the Nav2 tree handles route start acquisition and plan-and-follow execution.
- Behavior-Tree mission executive: The navigation loop plans over the next K=4 goals, prunes passed goals, and repeatedly shortens and re-smooths the controller path.These operations keep replanning frequent and local regardless of total route length.
- Layered recovery: Recovery escalates through costmap clearing, in-place rotation, waiting, and controlled backing under a bounded retry budget.Actions are attempted one at a time to refresh perception or free the robot when navigation becomes blocked.
- Reactive obstacle handling: The local costmap enforces reactive safety around sensed obstacles while the global costmap retains the operational boundary and known obstacles.Offline planning expresses route intent, whereas online perception handles safety during execution.
- Reactive obstacle handling and goal management: Goals exceeding cmax are deferred to the queue’s back, retried later, and dropped after rmax=3 postponements.Because the rolling local costmap can clear when obstacles disappear, blocked regions may be covered on later passes while reachable goals continue.
- Energy-aware mission management: Autonomous docking interrupts coverage below the battery threshold and first drives to a staging pose before perception-guided charging alignment.Return-to-charge is integrated into the same mission supervisor rather than handled as an external special case.
6 Evaluation
The integrated system was evaluated in simulation and across five outdoor areas spanning varied geometries, obstacles, and operating conditions. It completed the coverage missions autonomously while maintaining high executed coverage and handling docking and unknown obstacles.
- Experimental Setup: Five representative outdoor areas covered convex and non-convex geometries, with and without known obstacles, and reached about 1600 m2.Trials varied field shape and size, known-obstacle footprints, and unknown obstacles such as pedestrians or temporarily placed objects.
- Coverage Results: The representative Seq. 05 run followed planned parallel swaths across a convex field containing three obstacles.The comparison shows the planned route and executed swept area alongside mapped obstacle footprints.
- Mission Reliability: Every mission completed end-to-end without human intervention, including route execution and terminal docking when required.Mission success required reaching the route start, traversing the full coverage route, and completing the terminal phase.
- Coverage Results: 93.1% to 96.1% of the planned coverage area was swept across the five outdoor areas.Coverage efficiency uses the executed swept area inside the operational boundary divided by planned coverage, with a 1.0 m implement width.
- Coverage Results: Coverage quality remained largely insensitive to scale and clutter, with only three efficiency points separating the smallest convex field from the largest cluttered field.The largest field measured 1641 m2 and contained three mapped obstacles totaling 98.5 m2.
- Planning Efficiency: Planned routes used 1.15–1.18 meters of path per square meter of planned coverage across all five areas.This ratio remained essentially constant with field size, and offline planning completed within seconds even on the largest layout.
- Obstacle Handling: Unknown obstacles were detected online, causing the robot to slow, defer blocked goals, route around obstacles where possible, and resume coverage autonomously.Pedestrians walked alongside and in front of the robot during field trials, while the local costmap and goal-management layer maintained clearance.
7 Conclusion and Future Work
The paper presents a unified ROS 2 system for single-robot outdoor coverage and evaluates it across five outdoor areas with varied shapes, sizes, and obstacle layouts. Every route was completed while sweeping 93.1% to 96.1% of planned coverage, with seconds-scale offline planning and autonomous obstacle handling and docking; future work targets multi-robot coordination and localization under highly degraded or unavailable GNSS.
- Conclusion: The unified ROS 2 system combines dual-antenna RTK-GNSS localization, controller-oriented Fields2Cover refinements, and a Nav2 Behavior Tree executive.The executive supports long-duration execution with obstacle handling, recovery, and autonomous docking.
- Conclusion: 93.1% to 96.1% of planned coverage was swept across five outdoor areas, with every coverage route completed.Offline planning finished in seconds, and the executive handled unknown obstacles and returned to charge without intervention.
- Future Work: Future work will extend the system to multi-robot coordination and localization methods reliable when GNSS is highly degraded or unavailable.