Pseudo 3D LiDAR Development and Evaluation

In autonomous navigation, 2D LiDAR systems have inherent limitations when detecting objects with vertical variation, such as dock structures or overhanging obstacles. This experiment documents the development of a Pseudo 3D LiDAR by combining an RPLiDAR-A2M12 sensor with a mechanical tilting mechanism.

1. Motion Mechanism and Actuation

To obtain the third dimension ($Z$), the 2D LiDAR is mounted on a vertically actuated tilting platform.

• Motion Scheme: Continuous oscillating motion.
• Actuator: PowerHD 180° servo motor.
• Design Rationale: This approach was selected for its mechanical simplicity during early prototyping, as opposed to continuous rotation systems that require a slip ring.

Angle Synchronization

Synchronization between LiDAR range data and the servo elevation angle is achieved through PWM calibration:

• $0^{\circ}$ (Default Angle): Defined at PWM 1500.
• Minimum Angle: Measured at PWM 2600.
• Maximum Angle: Measured at PWM 700.

2. Coordinate Transformation (ROS Laser Assembler)

The RPLiDAR outputs 2D polar measurements ($d, \theta$). To construct a 3D point cloud, these measurements must be transformed using the servo elevation angle ($\phi$). This experiment employs the laser_assembler package under ROS Melodic.

Mathematically, the conversion to 3D Cartesian coordinates follows:

$$x = d \cdot \cos(\theta) \cdot \cos(\phi)$$
$$y = d \cdot \sin(\theta) \cdot \cos(\phi)$$
$$z = d \cdot \sin(\phi)$$

(refresh browser if equation does not appear)

Within the ROS ecosystem, this process is automated via the tf (Transform Tree), linking the vehicle’s base_link to the laser_frame through an intermediate servo_link.

3. Experimental Results and Performance Metrics

All experiments were conducted using an NVIDIA Jetson Nano for data processing and point cloud visualization.

Table 1. Technical Specifications and Performance of the Pseudo 3D LiDAR

4. Evaluation and Implementation Challenges

Although the system successfully produced a 3D representation of obstacles—particularly beneficial for detecting structures during docking maneuvers that are difficult to capture with pure 2D scans—several critical limitations were identified:

1. High Latency: A 7-second scan cycle is far too slow for dynamic ASV (Autonomous Surface Vehicle) navigation, significantly increasing collision risk at higher vessel speeds.
2. Data Stability: Without IMU integration to compensate for pitch and roll induced by wave motion, the reconstructed point cloud exhibits noticeable distortion and drift.
3. Sensor Limitations: The RPLiDAR-A2M12 relies on triangulation, whose performance degrades severely under direct sunlight, making it unsuitable for outdoor RoboBoat competition conditions.

5. Conclusions and Design Decision

This Pseudo 3D LiDAR experiment provides a valuable proof of concept, highlighting the importance of multi-dimensional spatial awareness in autonomous navigation. However, based on the evaluation results:

• The system was not adopted for the main competition due to its insufficient update rate.
• The team decided to await the deployment of native 3D LiDAR solutions (such as solid-state or multibeam LiDARs) offering significantly higher refresh rates.
• Current development efforts have shifted toward vision-based sensing (YOLOv4-Tiny) to improve detection of objects that are poorly perceived by 2D LiDAR alone.