AI Driving Olympics Home Challenges Submissions

Challenge "aido5-LFP-sim-validation"

Challenge description

LFP
Avoid the duckies pedestrians

In this variation of the lane following challenge, you must follow the lane and avoid the duckies.

See the documentation for more details .

Note: this is the validation challenge - you can see the output but these are not the scores that count.

Leaderboard

Submissions

Challenge logistics

Scoring

Scoring criteria

These are the metrics defined:

Survival time - survival_time_median

This is the median survival time. The simulation is terminated when the car goes outside of the road or it crashes with an obstacle or pedestrian.

Major infractions - in-drivable-lane_median

This is the median of the time spent outside of the drivable zones. For example this penalizes driving in the wrong lane.

Traveled distance - driven_lanedir_consec_median

This is the median distance traveled, along a lane. (That is, going in circles will not make this metric increase.)

This is discretized to tiles.

Lateral deviation - deviation-center-line_median

This is the median lateral deviation from the center line.

Dependencies

Dependencies

Depends on successful evaluation on LF 🚗 - Lane following (simulation 👾, validation 🏋)

The submission must first pass the LF validation.

The sum of the following tests should be at least 2.0.

Test on absolute scores:

good_enough(1.0 points)
Obtain at least 0.2 for score driven_lanedir_consec_median.

Test on relative performance:

better-than-bea-straight(1.0 points)
Do at least as good as a submission of 1639 labeled straight.

Details

Technical details

Evaluation steps details

Evaluation step LFP-sim

Timeout 10000.0

This is the Docker Compose configuration skeleton:

version: '3'
services:
    evaluator:
        image: docker.io/andreacensi/aido5-lfp-sim-validation-lfp-sim-evaluator:2020_12_07_10_32_32@sha256:28b18c8eb90b1c955e4a1ef7b8f488de1eeb991146231e13f62bd522e44a98d2
        environment:
            experiment_manager_parameters: 'episodes_per_scenario: 1

                episode_length_s: 60.0

                min_episode_length_s: 0.0

                seed: 20200922

                physics_dt: 0.05

                max_failures: 2

                fifo_dir: /fifos

                sim_in: /fifos/simulator-in

                sim_out: /fifos/simulator-out

                sm_in: /fifos/scenario_maker-in

                sm_out: /fifos/scenario_maker-out

                timeout_initialization: 120

                timeout_regular: 120

                port: 10123

                scenarios:

                - /scenarios

                '
        ports:
        - '10123'
    simulator:
        image: docker.io/duckietown/challenge-aido_lf-simulator-gym@sha256:d44a898930f18b99d8502b007a22a3db50ff1caf620eb8521eac83c335aaca27
        environment:
            AIDONODE_CONFIG: "env_constructor: Simulator\nenv_parameters:\n  max_steps:\
                \ 500001 # we don't want the gym to reset itself\n  domain_rand: 0\n\
                \  camera_width: 640\n  camera_height: 480\n  distortion: true\n \
                \ num_tris_distractors: 0\n  color_ground: [0, 0.3, 0] # green\n \
                \ enable_leds: true\n"
            AIDONODE_DATA_IN: /fifos/simulator-in
            AIDONODE_DATA_OUT: fifo:/fifos/simulator-out
    solution-ego0:
        image: SUBMISSION_CONTAINER
        environment:
            AIDONODE_NAME: ego0
            AIDONODE_DATA_IN: /fifos/ego0-in
            AIDO_REQUIRE_GPU: '1'
            AIDONODE_DATA_OUT: fifo:/fifos/ego0-out

The text SUBMISSION_CONTAINER will be replaced with the user containter.

Resources required for evaluating this step

Cloud simulations1