swarmt / go-voxblox

go-voxblox

A Go port(ish) of Voxblox modified to use sparse leaves and transmit mesh updates over gRPC as glTF (glb).

I wrote this to learn golang so some or all of it might make your eyes bleed.

System Diagram

graph LR
  tsdfIntegrator[TSDF Integrator]
  tsdfMap[TSDF Map]
  meshIntegrator[Mesh Integrator]
  meshLayer[Mesh Layer]

  Sensor -- Pointcloud2 --> tsdfIntegrator
  poseEstimate -- 6DOF Pose --> tsdfIntegrator
  tsdfIntegrator --> tsdfMap --> tsdfIntegrator
  tsdfMap -- Updated Blocks --> meshIntegrator
  meshLayer -- Set Updated False --> tsdfMap
  meshIntegrator --> meshLayer -. gRPC .-> glTF(glTF Mesh Blocks)

Test

go test go-voxblox

Run

Download the Cow and Lady Dataset and decompress (required for real time playback) with:

rosbag decompress data.bag

Settings are configured in voxblox.yaml.

The simple and fast integrators are available however the code runs the fast integrator by default.

Start a roscore with:

roscore

Start the node with:

go build go-voxblox

Optionally start the rerun viewer with:

cd viewer
python3 main.py

Start the bag playback with:

rosbag play --clock data.bag

You should see logs similar to the following:

2023/12/12 20:58:34 Convert PointCloud2: 2.312166ms
2023/12/12 20:58:34 Integrate Fast: 6.760125ms
2023/12/12 20:58:34 Convert PointCloud2: 3.882959ms
2023/12/12 20:58:34 Integrate Fast: 13.282292ms
2023/12/12 20:58:34 Integrate Mesh: 2.588875ms

Generate gRPC files

If you need to regenerate the protobuf and gRPC files you can do so with the following command:

protoc --go_out=. --go_opt=paths=source_relative \
--go-grpc_out=. --go-grpc_opt=paths=source_relative \
proto/mesh_service.proto 

TODO

• Merged integrator weights and speed
• Better / more unit tests
• Cache distant blocks with protobuf
• Logging
• System tests
• Stress test / map size
• ICP (kiss-icp?)
• Block updated status per subscriber

References

• CHISEL