Team I: General Hardware Agnostic Robot Software (GHAR)

Progress Review 4

Goals Mentioned:

• Full working of Kimera slam with semantics.
• Map conversion to planner-compatible map type
• Making our own traversability map from the elevation map
• Extending the planner to work with traversability maps

Planning

Things done so far

• Generated traversability map from elevation map, that considers height of obstacles, slope and roughness.
• Modified planner to work with traversability maps.

1. Planning

​

5

• Modified the random map generator to add obstacles with ramps
• Generated traversability map from the elevation map

Elevation map having obstacles with sloped edges

Traversability map of the same elevation map

1. Planning

​

6

Planner plans path through the slope that is gentler

1. Planning

​

7

Planner plans path over very low obstacles if the distance travelled is far less.

Things to be done in Planning

• More rigorous testing to make sure the planner works correctly
• Add interface for adding and removing obstacles
• Add roughness metric to cost function

SLAM

Things done so far

• Semantic Segmentation with pre-trained weights.

​

• Optimization using TensorRT (from 5 fps to 15 fps with MobileNet).

​

• Integrated segmented image with Kimera Semantics Successfully.

​

• Got a ply map of the MRSD Lab using Realsense D435i rosbag data.

2. SLAM

2. SLAM

​

11

• Kimera VIO and Kimera Semantics on Realsense 435i.
• Semantic segmentation is getting integrated.

Results

2. SLAM - Map Conversion

• Implementation Issue
• Elevation mapping not working with the map generated by SLAM

​

• Status: Still in Progress
• Not sure what the issue is, best guess so far is the map frame id.
• Need more testing

Future work to be done in SLAM

• Build bigger Maps
• Build maps of Level A, NSH.

​

• Better Semantic Segmentation model
• Train or use a better segmentation model for indoor data

​

• Perform Localization using the map generated.
• Use Kimera to do localization well and compare it with RTABMap

​

• Get Performance metrics.
• Satisfy the SLAM metrics from functional requirements.

Things to Do

• To Do
• Debug the semantics module fully and test with rosbags
• Continue with map conversion testing (decide if pivot is necessary)
• Look into semantic segmentation algorithms (replace voxblox)

Risk management and Issues Log

4. Risk Management

16

5. Issues Log

17

Thank You!

18