DustBotRobot Navigation and Obstacle Avoidance: A State-of-the-Art Review
Scope of This Review
This article reviews the state of the art in navigation and obstacle avoidance for autonomous outdoor service robots, with particular attention to the methods developed within and around the DustBot project (FP6-045299, 2006–2009). It is not a comprehensive survey of all mobile robot navigation — that literature spans thousands of papers — but rather a focused examination of the approaches most relevant to pavement-scale service robots operating in pedestrian environments.
The DustBot Navigation Architecture
DustCart employed a hierarchical navigation system with three layers:
- Topological planning — a graph-based representation of the road network, where nodes represent intersections and edges represent street segments. The global planner computed the shortest path through this graph from the robot’s current position to the requested destination.
- Metric navigation — within each street segment, the robot used differential GPS and laser rangefinder data to follow the planned trajectory with centimetre-level precision.
- Reactive obstacle avoidance — a local behaviour that overrode the metric navigation when sensors detected an obstacle within a safety threshold. The robot would decelerate, compute an avoidance manoeuvre, and resume its trajectory once the obstacle was cleared.
This three-layer architecture — plan, execute, react — is a standard pattern in mobile robotics, described in various forms by Arkin (1998), Siegwart and Nourbakhsh (2004), and others. DustBot’s contribution was demonstrating its effectiveness in unstructured outdoor environments with real pedestrian traffic.
Localisation Methods
DustCart’s primary localisation was differential GPS, achieving approximately 2 cm accuracy with a local base station. This was supplemented by laser-based feature matching for areas with degraded satellite coverage. The fusion of GPS and laser data used an Extended Kalman Filter (EKF), which maintains a probabilistic estimate of the robot’s position and updates it with each new sensor reading.
Mattoli et al. (2008) described the localisation system in detail, noting that the EKF maintained robust position estimates even when individual sensor modalities experienced temporary failures. The key was appropriate modelling of sensor noise covariances — GPS noise is different in character from laser noise, and the filter must account for this.
Obstacle Detection and Classification
DustBot’s 2D laser scanners provided reliable detection of objects at scan height (approximately 30–40 cm above ground). The system classified obstacles as either static (walls, bollards, parked vehicles) or dynamic (pedestrians, animals) based on temporal analysis — objects that appeared in one scan and were absent from the next were flagged as dynamic.
This binary classification was sufficient for basic avoidance but lacked the nuance that modern systems provide. Current approaches using 3D LiDAR and deep learning can classify obstacles into specific categories (pedestrian, cyclist, vehicle, animal) and predict their future trajectories, enabling more sophisticated avoidance strategies.
Path Planning Under Uncertainty
A significant challenge for outdoor service robots is planning paths when the environment is only partially known. Construction zones appear, market stalls are erected, vehicles park in unexpected locations. The DustBot system handled this through a combination of pre-mapped static environments and real-time sensor-based updates.
The D* Lite algorithm (Koenig and Likhachev, 2002) provides an efficient framework for replanning when the environment changes. While it is not confirmed whether DustBot used D* Lite specifically, the project’s publications describe an incremental replanning capability consistent with this family of algorithms.
Comparison with Contemporary Methods
| Aspect | DustBot (2007–2009) | Current Practice (2024–2026) |
|---|---|---|
| Localisation | DGPS + laser + EKF | RTK-GPS + LiDAR + visual SLAM + factor graphs |
| Obstacle detection | 2D laser, binary classification | 3D LiDAR + cameras, deep learning classification |
| Path planning | Graph-based global + reactive local | Semantic-aware planners, learned cost maps |
| Prediction | None (reactive only) | Trajectory prediction models (social force, deep learning) |
| Computing | Embedded PC | GPU-accelerated (Jetson, Orin) |
Key Publications
The following papers are directly relevant to DustBot’s navigation research:
- Mazzolai, B., Mattoli, V., Mondini, A., Laschi, C., Dario, P. (2008). “DustBot: a networked system of autonomous robots for urban hygiene.” CLAWAR 2008.
- Mattoli, V., Mondini, A., Mazzolai, B., Dario, P. (2008). “Navigation system design for an autonomous robot operating in urban outdoor environments.” IEEE/RSJ IROS 2008 Workshop on Planning, Perception and Navigation for Intelligent Vehicles.
- Lilienthal, A.J., Loutfi, A., Duckett, T. (2006). “Airborne chemical sensing with mobile robots.” Sensors, 6(11), 1616–1631.
- Koenig, S., Likhachev, M. (2002). “D* Lite.” AAAI Conference on Artificial Intelligence.