Alignment of data captured by autonomous vehicles to generate high definition maps
A high-definition map system receives sensor data from vehicles travelling along routes and combines the data to generate a high definition map for use in driving vehicles, for example, for guiding autonomous vehicles. A pose graph is built from the collected data, each pose representing location and orientation of a vehicle. The pose graph is optimized to minimize constraints between poses. Points associated with surface are assigned a confidence measure determined using a measure of hardness/softness of the surface. A machine-learning-based result filter detects bad alignment results and prevents them from being entered in the subsequent global pose optimization. The alignment framework is parallelizable for execution using a parallel/distributed architecture. Alignment hot spots are detected for further verification and improvement. The system supports incremental updates, thereby allowing refinements of sub-graphs for incrementally improving the high-definition map for keeping it up to date.
1. A method for generating high definition maps for use in the driving of autonomous vehicles, the method comprising:
receiving sensor data captured by a plurality of vehicles driving through a path in a geographical region;
generating a pose graph, wherein each node of the pose graph represents a pose of a vehicle, the pose comprising a location and orientation of the vehicle, and wherein each edge between a pair of nodes represents a transformation between nodes of the pair of nodes;
selecting a subset of nodes from the pose graph, wherein selecting the subset of nodes from the pose graph comprises:
identifying nodes having high quality global navigation satellite system (GNSS) poses; and
increasing the likelihood of selecting the identified nodes compared to nodes with lower quality GNSS poses;
for each node of the subset of nodes, identifying a GNSS pose corresponding to the node;
performing optimization of the pose graph based on constraints that minimize the pose difference between each of the nodes of the subset and the corresponding GNSS pose;
merging sensor data captured by plurality of autonomous vehicles to generate a point cloud representation of a geographical region;
generating a high-definition map based on the point cloud representation; and
sending the high-definition map to one or more autonomous vehicles for navigating in the geographical region.
2. The method of claim 1 , wherein the sensor data comprises LIDAR range images.
3. The method of claim 1 , further comprising:
performing pairwise alignment between pairs of nodes associated with a single track, wherein a track represents a collection of data corresponding to a single drive of a vehicle through a route, the data comprising a sequence of poses with associated sensor data.
4. The method of claim 3 , further comprising:
performing pairwise alignment between pairs of nodes, wherein a first node of the pair is selected from a first track and a second node of the pair is selected from a second track.
5. The method of claim 4 , further comprising:
performing filtering of results of pairwise alignments to determine whether a result needs further review.
6. The method of claim 5 , wherein the filtering of the results of pairwise alignments is performed using a machine learning based model configured to predict a score indicating whether a pairwise alignment result needs further review.
7. The method of claim 1 , further comprising:
removing one or more samples of sensor data responsive to determining that the samples were captured by a stationary vehicle.
8. The method of claim 1 , wherein selecting the subset of nodes from the pose graph comprises selecting nodes at fixed distance intervals along a path traveled by an autonomous vehicle capturing the sensor data corresponding to the node.
9. The method of claim 1 , wherein selecting the subset of nodes from the pose graph comprises a random selection of nodes from the pose graph.
10. The method of claim 1 , wherein selecting the subset of nodes from the pose graph comprises:
determining bounding boxes based on latitude and longitude values; and
selecting a node from each bounding box.
11. The method of claim 1 , wherein selecting the subset of nodes from the pose graph comprises:
identifying a region having low quality of pairwise alignment; and
increasing the sampling frequency in the identified region.
12. A non-transitory computer readable storage medium storing instructions for:
receiving sensor data captured by a plurality of vehicles driving through a path in a geographical region;
generating a pose graph, wherein each node of the pose graph represents a pose of a vehicle, the pose comprising a location and orientation of the vehicle, and wherein each edge between a pair of nodes represents a transformation between nodes of the pair of nodes;
selecting a subset of nodes from the pose graph, wherein selecting the subset of nodes from the pose graph comprises:
identifying nodes having high quality global navigation satellite system (GNSS) poses; and
increasing the likelihood of selecting the identified nodes compared to nodes with lower quality GNSS poses;
for each node of the subset of nodes, identifying a GNSS pose corresponding to the node;
performing optimization of the pose graph based on constraints that minimize the pose difference between each of the nodes of the subset and the corresponding GNSS pose;
merging sensor data captured by plurality of autonomous vehicles to generate a point cloud representation of a geographical region;
generating a high-definition map based on the point cloud representation; and
sending the high-definition map to one or more autonomous vehicles for navigating in the geographical region.
13. The non-transitory computer readable storage medium of claim 12 , wherein the stored instructions are for further:
performing pairwise alignment between pairs of nodes associated with a single track, wherein a track represents a collection of data corresponding to a single drive of a vehicle through a route, the data comprising a sequence of poses with associated sensor data.
14. The non-transitory computer readable storage medium of claim 13 , wherein the stored instructions are for further:
performing pairwise alignment between pairs of nodes, wherein a first node of the pair is selected from a first track and a second node of the pair is selected from a second track.
15. The non-transitory computer readable storage medium of claim 14 , wherein the stored instructions are for further:
performing filtering of results of pairwise alignments to determine whether a result needs further review, wherein the filtering of the results of pairwise alignments is performed using machine learning based model configured to predict a score indicating whether a pairwise alignment result needs further review.
16. The non-transitory computer readable storage medium of claim 12 , wherein the stored instructions are for further:
removing one or more samples of sensor data responsive to determining that the samples were captured by a stationary vehicle.
17. A computer system comprising:
an electronic processor; and
a non-transitory computer readable storage medium storing instructions executable by the electronic processor, the instructions for:
receiving sensor data captured by a plurality of vehicles driving through a path in a geographical region;
generating a pose graph, wherein each node of the pose graph represents a pose of a vehicle, the pose comprising a location and orientation of the vehicle, and wherein each edge between a pair of nodes represents a transformation between nodes of the pair of nodes;
selecting a subset of nodes from the pose graph, wherein selecting the subset of nodes from the pose graph comprises:
identifying nodes having high quality global navigation satellite system (GNSS) poses; and
increasing the likelihood of selecting the identified nodes compared to nodes with lower quality GNSS poses;
for each node of the subset of nodes, identifying a GNSS pose corresponding to the node;
performing optimization of the pose graph based on constraints that minimize the pose difference between each of the nodes of the subset and the corresponding GNSS pose;
merging sensor data captured by plurality of autonomous vehicles to generate a point cloud representation of a geographical region;
generating a high-definition map based on the point cloud representation; and
sending the high-definition map to one or more autonomous vehicles for navigating in the geographical region.
18. The computer system of claim 17 , wherein the stored instructions are for further:
performing pairwise alignment between pairs of nodes associated with a single track, wherein a track represents a collection of data corresponding to a single drive of a vehicle through a route, the data comprising a sequence of poses with associated sensor data.
19. The computer system of claim 18 , wherein the stored instructions are for further:
performing pairwise alignment between pairs of nodes, wherein a first node of the pair is selected from a first track and a second node of the pair is selected from a second track.