OSREC.018 answers a question every autonomous mobile robot must solve: Where am I on the map? A navigation plan is unreliable when its starting position is wrong. Adaptive Monte Carlo Localization (AMCL) estimates a robot’s position and heading on an existing map by comparing laser scans with expected surroundings. The method complements OSREC.015 on occupancy-grid mapping and OSREC.016 on A* pathfinding.
Learning objectives and prerequisites
- Define pose, particle, odometry, and likelihood.
- Explain the prediction, measurement, and resampling cycle.
- Distinguish the
map,odom, andbase_linkreference frames. - Recognize misleading localization and choose safe troubleshooting checks.
Prerequisites: coordinate frames, probability basics, 2D LiDAR scans, and OSREC.013 on robot state estimation. Exercises use recorded or simulated data; no physical robot movement is required.
How a robot represents uncertainty
A pose is the robot’s location and orientation, often represented in 2D by coordinates (x, y, yaw). Wheel encoders and inertial sensors estimate movement, but their errors accumulate. AMCL maintains many candidate poses called particles. Each particle represents a possible robot location, with a weight expressing how well measurements support it. A dense cluster signals a more concentrated hypothesis, not guaranteed correctness.
The estimator cycles through three operations: predict new particle poses from odometry and a motion-noise model; weight candidates by comparing incoming LiDAR ranges with a known occupancy map; and resample to concentrate candidates around plausible positions. Nav2 uses adaptive sampling to vary the particle population within configured limits. Consult its official AMCL parameter reference.
Map, odom, and base_link are different
The map frame describes the global map; odom provides smooth local motion that can drift; base_link tracks the robot body. AMCL normally supplies the map → odom correction, while an odometry source supplies odom → base_link. The combined transform locates the body on the map. An incorrect timestamp or two competing transform publishers can break navigation. ROS REP-105 defines the conventions.
A warehouse example
Imagine two parallel aisles with identical shelves. The robot starts with candidate poses in both aisles. As it moves past a distinctive loading bay, incoming LiDAR measurements may increasingly favor one candidate cluster. A doorway blocked by a temporary pallet can instead make the scan disagree with the static map. AMCL may downweight misleading candidates, but widespread changes or repetitive geometry can still create ambiguity. Localization estimates uncertainty; it does not guarantee obstacle avoidance.
Safe validation workflow
- Load the correct, version-matched occupancy map in the simulator and check the LiDAR topic and sensor-frame transform.
- Set a plausible initial pose and heading using the supported Nav2/RViz interface; inspect the resulting particle distribution.
- Compare observed scan endpoints against mapped walls as the simulated robot is moved under controlled conditions.
- Check that only the designated localization process publishes
map → odom; inspect timestamp alignment and odometry continuity. - Document the pose estimate, uncertainty, map version, and failures before authorizing navigation. Stop motion on repeated pose jumps.
Illustrative configuration: Nav2’s Jazzy AMCL reference lists defaults of min_particles: 500 and max_particles: 2000. The permitted span is 1,500 particles, not a measurement of guaranteed localization accuracy. Higher counts can consume additional computing resources. Settings depend on the map, sensors, motion model, and release.
Practice and answer key
Exercise: A simulated robot’s odometry remains smooth, but its plotted map position jumps between similar aisles. Name two possible causes and one safe next check.
Answer: Repeated geometric patterns or an incorrect initial pose can produce competing hypotheses; a wrong LiDAR-to-base transform or stale map is another possibility. Inspect scan-to-map alignment and the transform tree before resuming autonomous travel.
Knowledge check: Does AMCL construct the map? Answer: No. It normally localizes against an existing map. Contrast the mapping work in OSREC.014 on pose-graph SLAM and the path repairs in OSREC.017 on D* Lite. The principal safety rule is to validate localization before trusting a route.
We volunteer daily to ensure the credibility of the information on this platform is Verifiably True. If you would like to support to help further secure the integrity of our research initiative, please scan the QR code below.


Leave a Reply