Saronic State Prediction Challenge

Context

Vehicle dynamics on a robot in the water can be challenging. This challenge is a first pass on understanding the challenges behind how vehicle controls in varying sea states can be difficult.

Goal

We want two predictors:

• Predict the next vehicle state given a sequence of vehicle controls and the current vehicle state.

  • vehicle state is an AHRS message, see the ahrs.log for an example

• Predict the sequence of vehicle controls to achieve a desired vehicle state given a current vehicle state

  • vehicle state is an AHRS message, see the ahrs.log for an example

Dataset

We have provided two dataset files which can be fetched using git lfs pull after installing git lfs:

• vehicle_control.csv

  • This is a csv with ts, throttle and turn values
    • ts is a timestamp in UTC
    • throttle is a float value between 0.0-1.0
    • turn is a float value between -1.0 and 1.0
      • 0.0 to 1.0 is a right hand turn
      • -1.0 to 0.0 is a left hand turn
    • gear is a discrete motor setting:
      • 0 neutral
      • 1 forward
      • 2 reverse
    • trim is the discrete trim command that affects boat dynamics, taking on vals 0, 1 or 2.

• ahrs.csv

  • This is a csv with ts, roll_deg, pitch_deg, yaw_deg, ve_mps, vn_mps, vu_mps, omega_x_dps, omega_y_dps, omega_z_dps, ax_mps2, ay_mps2, az_mps2
    • ts is a timestamp in UTC
    • roll_deg, pitch_deg, and yaw_deg are degrees values
    • ve_mps, vn_mps, and vu_mps are velocities in meters per second in the ENU frame
    • omega_x_dps, omega_y_dps, and omega_z_dps are angular rates in the FRD frame
    • ax_mps2, ay_mps2, and az_mps2 are meter per second^2 accelerations in FRD frame

Submission

• Send us a git repo with code for the two predictors and instructions on how to run it
• Also add a README/document with a description explaining the design for your predictor
• Any visualizations or plots to accompany your code are encouraged in your document that explains your design