Skip to content

Latest commit

 

History

History
75 lines (51 loc) · 3.84 KB

README.md

File metadata and controls

75 lines (51 loc) · 3.84 KB

comma Controls Challenge v2

Prize money: best score between August 5th and September 1st 2024 wins $500!

Machine learning models can drive cars, paint beautiful pictures and write passable rap. But they famously suck at doing low level controls. Your goal is to write a good controller. This repo contains a model that simulates the lateral movement of a car, given steering commands. The goal is to drive this "car" well for a given desired trajectory.

Getting Started

We'll be using a synthetic dataset based on the comma-steering-control dataset for this challenge. These are actual car and road states from openpilot users.

# install required packages
# recommended python==3.11
pip install -r requirements.txt

# test this works
python tinyphysics.py --model_path ./models/tinyphysics.onnx --data_path ./data/00000.csv --debug --controller pid 

There are some other scripts to help you get aggregate metrics:

# batch Metrics of a controller on lots of routes
python tinyphysics.py --model_path ./models/tinyphysics.onnx --data_path ./data --num_segs 100 --controller pid

# generate a report comparing two controllers
python eval.py --model_path ./models/tinyphysics.onnx --data_path ./data --num_segs 100 --test_controller pid --baseline_controller zero

You can also use the notebook at experiment.ipynb for exploration.

TinyPhysics

This is a "simulated car" that has been trained to mimic a very simple physics model (bicycle model) based simulator, given realistic driving noise. It is an autoregressive model similar to ML Controls Sim in architecture. Its inputs are the car velocity (v_ego), forward acceleration (a_ego), lateral acceleration due to road roll (road_lataccel), current car lateral acceleration (current_lataccel), and a steer input (steer_action), then it predicts the resultant lateral acceleration of the car.

Controllers

Your controller should implement a new controller. This controller can be passed as an arg to run in-loop in the simulator to autoregressively predict the car's response.

Evaluation

Each rollout will result in 2 costs:

  • lataccel_cost: $\dfrac{\Sigma(actual\_lat\_accel - target\_lat\_accel)^2}{steps} * 100$

  • jerk_cost: $\dfrac{\Sigma((actual\_lat\_accel_t - actual\_lat\_accel_{t-1}) / \Delta t)^2}{steps - 1} * 100$

It is important to minimize both costs. total_cost: $(lataccel\_cost * 50) + jerk\_cost$

Submission

Run the following command, then submit report.html and your code to this form.

python eval.py --model_path ./models/tinyphysics.onnx --data_path ./data --num_segs 5000 --test_controller <insert your controller name> --baseline_controller pid

Changelog

  • With this commit we made the simulator more robust to outlier actions and changed the cost landscape to incentivize more aggressive and interesting solutions.
  • With this commit we fixed a bug that caused the simulator model to be initialized wrong.

Work at comma

Like this sort of stuff? You might want to work at comma! comma.ai/jobs