Description

The simulator's IMU sensors were not being populated with realistic values, causing issues with:

• locationd calibration
• paramsd steering angle offset learning
• overall vehicle control accuracy

This led to:

• locationd receiving zero values for sensor data
• paramsd learning incorrect steering offsets

Proposed Solution & Implementation

Implemented IMU sensor value calculation using vehicle state data from MetaDrive. The solution:

• Added IMU calculation logic in read_sensors()

• Added IMU calculation function that:

• Computes realistic accelerometer values considering:

  • Linear acceleration
  • Centripetal forces during turns
  • Gravitational acceleration (9.81 m/s²)

• Calculates gyroscope values from bearing changes

• Applies appropriate coordinate transformations

• Enforces realistic physical limits on sensor values

Testing and Validation

• For testing , wrote custom unit tests to ensure consistency , robustness and functionality .
• Test file can be currently found here : https://github.com/sambhavnoobcoder/openpilot/blob/Testing-Sensors-Bounty/tools/sim/tests/test_metadrive_world.py .
  NOTE: This file was not added in this PR because -
  a) This appeared out of scope of the bounty
  b) more importantly , it made the PR big in LOC . will add this this as a separate PR if required .
• The implementation:

1. Successfully generates non-zero, physically realistic IMU values
2. Maintains accelerometer readings within vehicle limits (<18 m/s²)
3. Produces reasonable gyroscope values during turns
4. Enables proper locationd calibration
5. Allows paramsd to learn accurate steering offsets
6. Improves overall steering angle correction

Sample output and Inference

Here is a small screen recording of the outputs generated:

This output was generated with the parameters :

metadrive_vehicle_state(
    velocity=vec3(x=3.25, y=10.50, z=0), 
    position=(217.06, 82.06),
    bearing=72.51,
    steering_angle=7.87
)

The inference drawn from this output was :

• Accelerometer shows expected acceleration patterns during turns
• Gyroscope correctly reflects vehicle rotation rate
• Calculated values enable proper steering offset learning (previous invalid 0° → now realistic ~2.7°)

Impact

This fix:

• Enables proper locationd calibration in simulator
• Improves steering angle offset learning
• Enhances simulation accuracy

The implementation maintains clean code organization while adding essential simulator functionality, making the simulation environment more realistic and useful for testing vehicle control systems.
Related Issues

• Fixes simulator: sensors aren't filled #33721
• Unblocks op sim: fix car steering angle #33718