SLAM

Benchmark Results for SLAM

FMCW LiDAR SLAM vs 4D Radar SLAM vs Spinning Radar SLAM

0. Baseline Method

We utilize three methods: we use Fast-LIO for FMCW LiDAR SLAM, 4DRadarSLAM for 4D Radar SLAM, and ORORA for Spinning Radar SLAM.
Absolute Trajectory Error (ATE) is measured in meters, while Relative Pose Error (RPE) is quantified in degrees per meter for rotation (RPEr) and as a percentage for translation (RPEt).

1. ATE and RPE result (Sports Complex, Library)

2. Estimated odometry result (Sports Complex, Library)

Sports Complex

Library

3. Mapping result of ORORA (Sports Complex)

4. SLAM result of HeRCULES

Among these baselines, the odometry result of Fast-LIO was the most accurate, followed by ORORA and 4DRadarSLAM. Using Spinning radar, the scale differs, but the ground truth and approximate trajectory are similar.
The result using 4D radar is not as good, but there is potential for improvement through preprocessing the raw point cloud to reduce noise.
These findings validate that SLAM performance with 4D radar alone is limited on our dataset, highlighting the need for heterogeneous radar SLAM or radar-LiDAR fusion SLAM.

5. Challenges - River Island

Various Dynamic Objects: Moving vehicles and pedestrians create constantly changing environments, making it hard for SLAM systems to differentiate between static landmarks and dynamic objects, leading to mapping and localization errors.
Intersections: Intersections lack distinct landmarks and involve multiple routes, complicating trajectory tracking and causing potential confusion in the system, especially when road markings are unclear.
8-Lane Road: Wide, featureless roads provide fewer landmarks, increasing the risk of localization drift. Managing traffic and lane changes adds further complexity, requiring real-time adjustments in the map and position estimates.

6. Challenges - Library

Uphill and Downhill: Changes in elevation can impact sensor accuracy and distort depth perception, making it challenging for systems to maintain precise localization and mapping. Inclines can also affect wheel odometry, further complicating position tracking.
Steep Curves: Sharp turns reduce the visibility of distant landmarks and can cause sudden changes in the robot's orientation. This makes it harder for SLAM systems to maintain an accurate map and track the robot's movement through these tight spaces.
Narrow Roads: In confined spaces, SLAM systems have limited room for error. With fewer visible landmarks and reduced maneuverability, the margin for accurate localization becomes tighter, making it difficult to avoid obstacles and maintain a reliable map.

Page updated

Google Sites

Report abuse