Skip to content

training.md

Latest commit

 

History

History
120 lines (89 loc) · 3.52 KB

training.md

File metadata and controls

120 lines (89 loc) · 3.52 KB

Training

Running a training session

docker run -d --name tal-train -p 6080:6080 -v tal-data:/app/Data tal-racing

Open http://localhost:6080/vnc.html to watch the agent learn in real time. Training logs stream to the container output:

12450::Lap Complete 18 -> FinalR: 1.00 -> LapTime 61.4 -> TotalReward: 99.2 -> Progress: 1.00
12510::Crashed -> FinalR: -1.00 -> LapTime 5.3 -> TotalReward: 4.1 -> Progress: 0.08

Check progress at any time:

docker logs tal-train 2>&1 | grep -E "Lap Complete|Crashed|Finished Training"

Config file structure

Config files live in config/. Each file defines a sweep of training runs.

test_name: "TAL_speeds"       # used as the save directory name

architecture: "fast"          # network architecture (fast = small MLP)
n_scans: 2                    # number of LIDAR scan subsets used as input
train_mode: "Std"
test_mode: "Std"

n: 4                          # number of random seeds to run
set_n: 1                      # set identifier (used in run name)
random_seed: 10000            # base seed; actual seed = random_seed + 10*n

noise_std: 0.1                # Gaussian position noise (metres), 0 to disable
lidar_noise_std: 0.05         # Gaussian LIDAR noise (metres), 0 to disable

n_train_steps: 100000         # training steps per run
n_test_laps: 20               # evaluation laps after training

map_name: "f1_esp"            # fixed map (used when runs vary speed)
reward: "TAL"                 # TAL | Cth | Progress

runs:                         # list of variable parameters across runs
  - max_speed: 4
  - max_speed: 5
  - max_speed: 6
  - max_speed: 7

To vary maps instead of speeds:

max_speed: 6
runs:
  - map_name: "f1_esp"
  - map_name: "f1_mco"
  - map_name: "f1_aut"
  - map_name: "f1_gbr"

Reward functions

Value Description
TAL Trajectory-Aided Learning — penalises deviation from the optimal racing line. Best overall performance.
Cth Cross-track heading reward — penalises lateral error and heading error from the racing line.
Progress Pure progress along the track. Simple but struggles at high speeds.

Available maps

ID Circuit
f1_esp Circuit de Barcelona-Catalunya (Spain)
f1_mco Circuit de Monaco
f1_aut Red Bull Ring (Austria)
f1_gbr Silverstone (Great Britain)

Trained model paths

Weights are saved to the Docker volume at:

Data/Vehicles/<test_name>/<run_name>/<run_name>_actor.pth
Data/Vehicles/<test_name>/<run_name>/<run_name>_critic.pth

The run name encodes all parameters:

fast_Std_Std_TAL_f1_esp_6_5_2
 │    │    │   │   │   │ │ └─ seed (n=2)
 │    │    │   │   │   │ └─── set_n
 │    │    │   │   │   └───── max_speed
 │    │    │   │   └───────── map
 │    │    │   └───────────── reward
 │    │    └───────────────── test_mode
 │    └────────────────────── train_mode
 └─────────────────────────── architecture

Extended training (all configs)

A start_train_extended.sh script runs all configs sequentially:

docker run -d --name tal-train-ext -p 6080:6080 -v tal-data:/app/Data tal-racing /start_train_extended.sh

This trains TAL_speeds (16 runs) then TAL_maps (20 runs) — approximately 18–20 hours on CPU.