SLAM Quickstart
This guide provides examples of using the SLAM API for development purposes. Users can run SLAM with an OS sensor’s hostname or IP for real-time processing, or with a recorded PCAP/OSF file for offline processing.
Warning
Due to missing upstream dependency support on python3.12, slam does not work on python3.12.
Obtain Lidar Pose and Calculate Pose Difference
The SLAM API outputs the sensor’s pose for each lidar scan, which you can use to determine the sensor’s orientation in your system. From the Lidar Poses, we can calculate the Pose difference between consecutive scans
from ouster.sdk import open_source
from ouster.mapping.slam import KissBackend
import numpy as np
scans = open_source(pcap_path, sensor_idx=0)
slam = KissBackend(scans.metadata, max_range=75, min_range=1, voxel_size=1.0)
last_scan_pose = np.eye(4)
for idx, scan in enumerate(scans):
scan_w_poses = slam.update(scan)
col = client.first_valid_column(scan_w_poses)
# scan_w_poses.pose is a list where each pose represents a column points' pose.
# use the first valid scan's column pose as the scan pose
scan_pose = scan_w_poses.pose[col]
print(f"idx = {idx} and Scan Pose {scan_pose}")
# calculate the inverse transformation of the last scan pose
inverse_last = np.linalg.inv(last_scan_pose)
# calculate the pose difference by matrix multiplication
pose_diff = np.dot(inverse_last, scan_pose)
# extract rotation and translation
rotation_diff = pose_diff[:3, :3]
translation_diff = pose_diff[:3, 3]
print(f"idx = {idx} and Rotation Difference: {rotation_diff}, "
f"Translation Difference: {translation_diff}")
SLAM with Visulizer and Accumulated Scans
Visualizers and Accumulated Scans are also available for monitoring the performance of the algorithm, as well as for demonstration and feedback purposes.
from functools import partial
from ouster.viz import SimpleViz, ScansAccumulator
from ouster.mapping.slam import KissBackend
scans = open_source(pcap_path, sensor_idx=0)
slam = KissBackend(scans.metadata, max_range=75, min_range=1, voxel_size=1.0)
scans_w_poses = map(partial(slam.update), scans)
scans_acc = ScansAccumulator(info,
accum_max_num=10,
accum_min_dist_num=1,
map_enabled=True,
map_select_ratio=0.01)
SimpleViz(info, scans_accum=scans_acc, rate=0.0).run(scans_w_poses)
More details about the visualizer and accumulated scans can be found at the Ouster Visualizer and Scans Accumulator
Note
The performance of the SLAM algorithm depends on your CPU’s processing power and the ‘voxel_size’ parameter. Below is a suggestion for selecting an appropriate voxel size:
Intro SLAM in Ouster-CLI
We also offer a simpler method to run SLAM using the ouster-cli. For additional details, please refer to Ouster-CLI Mapping.