What Is a 3D LiDAR and When Should You Use It Instead of 2D?
Adrian Krzemiński,
TL;DR: A 3D LiDAR captures multiple scan layers (often 16 to 128 in mechanical multi-beam models, or a non-repeating 3D scan pattern in some solid-state models) and returns a volumetric point cloud, while a 2D LiDAR samples a single horizontal plane. Choose 3D LiDAR when your mobile robot operates in unstructured terrain, needs to detect overhanging or low-profile obstacles, or runs 6-DoF SLAM. Stick with 2D LiDAR when you map flat indoor environments, need low cost and low CPU load, or only require planar obstacle avoidance. The decision comes down to whether your robot’s environment is geometrically flat in the sensor plane, or not.
What a 3D LiDAR Is and How It Differs From 2D
A LiDAR (Light Detection and Ranging) sensor measures distance by emitting laser pulses and timing their return. A 2D LiDAR spins a single laser emitter or mirror horizontally, producing one ring of range measurements per revolution. A 3D LiDAR stacks multiple emitters at different vertical angles, or steers a beam across two axes, producing a dense point cloud that describes the world in X, Y, and Z.
The practical difference is dimensionality. A 2D LiDAR sees a slice. A 3D LiDAR sees a volume. For a wheeled robot on a perfectly flat warehouse floor, a slice is often enough. For a UGV crossing a gravel path with low branches, a tree stump, or a curb, the slice can miss critical geometry entirely.
Modern 3D sensing setups used on robots usually fall into these groups:
- Mechanical spinning multi-beam (e.g. Velodyne-style 16/32/64-channel units): wide 360° horizontal FoV, moderate vertical FoV (typically 30-40°).
- Solid-state and MEMS (e.g. Livox, RoboSense M-series): fixed or electronically/MEMS-scanned FoV, no continuously rotating external housing, lower cost in many cases.
- 2.5D tilting or nodding 2D setups (e.g. a 2D scanner on a tilt mechanism): not true 3D LiDAR, but able to recover 3D structure over time.
How 2D LiDAR and 3D LiDAR Compare in Practice
The table below summarizes the trade-offs that matter most when integrating either sensor on a mobile platform such as the Leo Rover or Raph Rover.
| Property | 2D LiDAR | 3D LiDAR |
|---|---|---|
| Scan layers | 1 | 16-128 rings (mechanical), or non-repeating scan patterns / dense FoV samples (some solid-state) |
| Output rate | Tens of thousands of points/s | Hundreds of thousands to millions of points/s |
| Vertical FoV | 0° (planar) | Approximately 20-90° depending on model |
| Typical range | 5-40 m | 30-300 m depending on model and target reflectivity |
| Price range (approx.) | USD 100-1,500 | USD 600-10,000+ |
| CPU / GPU load | Low; runs on Raspberry Pi-class hardware | Moderate to high; often needs an x86 or Jetson-class compute |
| Typical SLAM | Gmapping, Cartographer 2D, slam_toolbox | LIO-SAM, FAST-LIO2, Cartographer 3D |
| Best for | Indoor AMRs, flat-floor navigation, docking | Outdoor UGVs, inspection, forestry, construction, autonomous driving research |
| Weakness | Blind to anything outside the scan plane | Cost, power draw, data volume |
When You Should Use 3D LiDAR Instead of 2D
The decision is rarely about resolution alone. It is about whether the missing dimension contains information your robot needs to act on. Use this flow to decide:
- Is the robot’s environment geometrically flat in the LiDAR plane? If yes (indoor warehouse, lab, office), a 2D LiDAR is usually sufficient.
- Are there obstacles below or above the scan plane? Forklift tines, table edges, low curbs, tree branches, cables. If yes, you need 3D LiDAR or a sensor fusion stack with depth cameras.
- Does the robot pitch or roll significantly? Field robots traversing slopes will tilt their 2D scan plane into the ground or sky. 3D LiDAR is more robust to this; 2D requires IMU-based compensation or a stabilized mount, with limits.
- Do you need 6-DoF localization? LIO-SAM, FAST-LIO2, and similar pipelines require multi-beam or true 3D LiDAR input. 2D SLAM gives you 3-DoF (x, y, yaw) only.
- Is the deliverable a 3D map? Inspection, digital twins, forestry inventory, BIM verification – all require volumetric data.
If you answered “yes” to any of items 2-5, you are looking at 3D LiDAR territory. If all answers were “no,” 2D LiDAR will save you cost, power, and integration time.
How 3D LiDAR Performs on Field Robots Like Leo Rover
Field robotics is where the 2D versus 3D trade-off becomes concrete. On a Leo Rover navigating outdoor terrain, a horizontally mounted 2D LiDAR sees a ring that constantly tilts as the chassis pitches over rocks and roots. The result is false positives (ground hits read as obstacles) and false negatives (low-profile hazards missed between scans).
A 3D LiDAR resolves this in two ways. First, the vertical FoV intersects both ground and obstacles, so ground segmentation algorithms (e.g. Patchwork++, ray ground filter) can separate traversable from non-traversable surfaces. Second, point cloud registration via ICP, NDT, or LiDAR-inertial odometry can provide odometry independent of wheel encoders when they slip on loose surfaces. This is why research applications in outdoor autonomy, agricultural robotics, and SLAM benchmarking often specify 3D LiDAR.
For inspection tasks, 3D LiDAR enables direct deliverables: a registered point cloud of a structure, a stockpile volume measurement, or a clearance check. A 2D LiDAR cannot produce these outputs without external motion (e.g. mounting it on a pan-tilt or relying on robot motion to “push-broom” scans together).
Why 2D LiDAR Still Matters
The temptation is to assume 3D is strictly better. It is not. 2D LiDARs remain the default for several reasons:
- Cost. A capable 2D unit like the Slamtec RPLIDAR A-series costs under USD 400. A 16-beam 3D LiDAR commonly costs around USD 1,000 or more, although entry-level and promotional pricing can be lower.
- Compute budget. 2D scans are tiny. They process on a Raspberry Pi 4 with headroom. 3D point clouds at 10-20 Hz often benefit from a Jetson Orin or comparable compute.
- Mature ROS 2 stack. Nav2 with slam_toolbox is production-ready for 2D. 3D SLAM in ROS 2 is improving but more fragmented.
- Power. 2D LiDARs draw 2-5 W. Multi-beam 3D units can pull 8-25 W, which matters on battery-powered UGVs.
A 2D sensor like the Slamtec RPLIDAR S3 sits on the lower-cost side of this trade-off: it is a long-range (40 m) 2D scanner that, when mounted on a tilting mechanism or combined with robot motion, can support 2.5D or 3D mapping workflows for some practical outdoor use cases without the cost of a multi-beam unit.
How 3D LiDAR Integrates With ROS 2
On the software side, 3D LiDAR drivers publish sensor_msgs/PointCloud2, often at 5-20 Hz. From there, common ROS 2 pipelines include:
- FAST-LIO2 / LIO-SAM for tightly-coupled LiDAR-inertial odometry. These require a synchronized IMU and produce drift-resistant 6-DoF poses.
- Cartographer 3D for offline or online mapping with loop closure, typically via ROS 1, bridges, or community ROS 2 ports.
- Nav2 with voxel-based obstacle layers using voxel layers or the spatio_temporal_voxel_layer plugin to feed 2D navigation costmaps from 3D data.
- Patchwork++ or linefit_ground_segmentation for ground extraction before planning.
Plan compute accordingly. FAST-LIO2 on a 16-beam scan runs comfortably on a Jetson Orin Nano. Cartographer 3D with loop closure benefits from x86. These workloads are generally not recommended on a Pi 4 for real-time use.
FAQ
Is a 3D LiDAR always better than a 2D LiDAR?
No. For flat-floor indoor navigation, a 2D LiDAR delivers equivalent navigation performance at a fraction of the cost and compute. 3D LiDAR becomes necessary when geometry outside the scan plane carries information the robot must use.
Can you do SLAM with a 2D LiDAR outdoors?
Yes, but with caveats. Open outdoor spaces often lack the vertical features 2D SLAM relies on, and robot pitch tilts the scan plane unpredictably. Results vary from acceptable on flat parking lots to unusable on rough terrain.
How much does a 3D LiDAR cost?
Entry-level solid-state units (e.g. Livox Mid-360) start around USD 600-1,500 depending on reseller, region, and taxes. Mechanical 16-beam units sit roughly in the USD 1,000-4,000 range. High-channel-count automotive-grade sensors exceed USD 8,000. Prices change frequently; check vendor listings for current figures.
What is the difference between a 3D LiDAR and a depth camera?
A depth camera (stereo, structured light, or ToF) produces dense depth over a narrow FoV at short range; active depth cameras in particular can degrade in sunlight. A 3D LiDAR provides sparser but longer-range (often 40 m+) measurements that work outdoors. They are often combined.
Do you need an IMU with a 3D LiDAR?
For modern LiDAR-inertial odometry (FAST-LIO2, LIO-SAM), yes – a 6-axis or 9-axis IMU synchronized with the LiDAR is required. For pure scan matching, no, but performance suffers in feature-poor environments.
Can the Leo Rover carry a 3D LiDAR?
Yes. The Leo Rover can carry many compact 3D LiDARs, but you should verify the payload, mounting, voltage, and power budget for the specific model. Compute is often the bigger constraint: pair the unit with a Jetson-class single-board computer rather than relying only on the stock Raspberry Pi for real-time 3D SLAM.
What is a “2.5D” LiDAR setup?
It refers to a 2D LiDAR mounted on a tilting or nodding mechanism, or paired with robot motion, to reconstruct 3D structure over time. Cheaper than true 3D LiDAR but slower and motion-dependent.
Browse the Slamtec RPLIDAR S3 in the Fictionlab shop to see specifications, ROS 2 driver support, and integration notes for Leo Rover and Raph Rover platforms.