Issues log
| Sl No | Date Found | Date Resolved | Owner | Origin | Description | Resolution | Artifacts Changed |
|---|---|---|---|---|---|---|---|
| 1 | 03/01/2020 | 03/25/2020 | Kartik | RL Node Integration | Needed to figure out a way to build a synchronous interface between an RL library and the existing framework | Implemented threading witch locks to achieve this | RL ROS node |
| 2 | 03/17/2020 | 03/17/2020 | Kartik | Carla Simulator | Issues building latest Carla Version
Function definition mismatches in Unreal source code as the wrong version of Unreal was mentioned in the installation steps. |
Temporarily changed the function definitions in the Unreal engine’s header files and re-compiled | Unreal Engine version changed |
| 3 | 03/17/2020 | 03/19/2020 | Scott | Carla Scenario Runner | Failed to get stable speed performance of the generated vehicles
The scenario runner built-in pytree’s chained behavior tree has this behavior: following leading vehicle changing the agent speed without considering collision . |
Added the collision check as a post-processing unit to change agent speed in order to avoid collisions. | Simulator subsystem structure changed |
| 4 | 03/19/2020 | 04/02/2020 | Alvin | System | One iteration of the system takes too long to compute | Combined path planner and behavior planner nodes together to avoid ROS communication between the nodes | RL ROS Node & Path Planner Node |
| 5 | 03/25/2020 | 03/26/2020 | Mayank | CARLA | Can’t spawn vehicles with non-zero initial speed in CARLA reliably | Designed the scenarios to begin with all vehicles on autopilot, and control of ego vehicle is given to our system after a fixed scenario running duration. |
Simulation Subsystem |
| 6 | 09/10/2020 | 09/16/2020 | Mayank | CARLA | Frenet coordinates have incorrect values in certain intersections, and slow. | Redesigned the frenet coordinate calculation to be able to handle intersections which may not be in perfect right angles. |
Simulation Subsystem |
| 7 | 10/15/2020 | 10/20/2020 | Kartik | State Manager Class in RL Node | Agent not learning to handle even simple scenarios. | Fixed the bug in creation of the state representation being fed to the neural net from the environment description message. |
RL ROS Node |
| 8 | 11/02/2020 | 11/10/2020 | Alvin | Point-to-Point Node | Agent crashing into vehicles for simple scenarios as well | Fixed the hand over conditions used to transition between different RL agents to give each agent the same conditions as it was trained in. |
Point-to-Point ROS Node |