Datasets:

harryzhang1018
/

TestRandomization

"""Fixed TR design for the held-out evaluation on goal set B (Sim2Sim).
Fixed before any goal-set-B run and before any evaluation of the held-out policies (seeds 2001-2004).
Chosen from the goal-set-A exploration (see variants.py and results/exploration_report.md).
PRIMARY_TR is the only condition in the primary test. All others are secondary and are reported whatever
their result. Distributions use the same parameter names as rom_batch.DEFAULTS; one trial = one scenario.
Condition names also seed the random streams (run_rom.seed_of), so they are kept unchanged.
"""
from variants import REL

TRIALS = 20

PRIMARY_TR = 'tr_rate'
BASELINE = 'nominal'

CONDITIONS = {
    # baseline: nominal ROM = the training ROM, deterministic, K = 1
    'nominal': (1, {}),
    # PRIMARY TR: noise-only TR (action N(0,0.05) on throttle and steering; tau0 N(1000,50), omega0 N(2600,100))
    # plus a random steering-rate limit, LogUniform(0.05, 0.5) normalized units per 0.1 s control step, one draw per trial
    'tr_rate': (TRIALS, {**REL, 'steer_rate': ('loguniform', 0.05, 0.5)}),
    # secondary: noise-only TR (action and powertrain noise, no rate limit)
    'paper_tr': (TRIALS, REL),
    # secondary: sensitivity of the primary TR to the range and the shape of the rate-limit distribution
    'tr_rate_half': (TRIALS, {**REL, 'steer_rate': ('loguniform', 0.025, 0.25)}),
    'tr_rate_double': (TRIALS, {**REL, 'steer_rate': ('loguniform', 0.1, 1.0)}),
    'tr_rate_uniform': (TRIALS, {**REL, 'steer_rate': ('uniform', 0.05, 0.5)}),
    # secondary: the other actuator family (random delay of 0-3 control steps) with the noise-only TR
    'tr_delay': (TRIALS, {**REL, 'delay': ('choice', [0, 1, 2, 3])}),
    # secondary deterministic baselines (not TR): the release Chrono loop's fixed rate limit 0.1;
    # the fixed rate limit with the 1 s start hold and the HMMWV wheelbase (3.378 m)
    'det_rate_0.1': (1, {'steer_rate': ('const', 0.1)}),
    'det_calibrated': (1, {'wheelbase': ('const', 3.378), 'steer_rate': ('const', 0.1), 'hold': ('const', 10)}),
}

	"""Fixed TR design for the held-out evaluation on goal set B (Sim2Sim).

	Fixed before any goal-set-B run and before any evaluation of the held-out policies (seeds 2001-2004).
	Chosen from the goal-set-A exploration (see variants.py and results/exploration_report.md).
	PRIMARY_TR is the only condition in the primary test. All others are secondary and are reported whatever
	their result. Distributions use the same parameter names as rom_batch.DEFAULTS; one trial = one scenario.
	Condition names also seed the random streams (run_rom.seed_of), so they are kept unchanged.
	"""
	from variants import REL

	TRIALS = 20

	PRIMARY_TR = 'tr_rate'
	BASELINE = 'nominal'

	CONDITIONS = {
	# baseline: nominal ROM = the training ROM, deterministic, K = 1
	'nominal': (1, {}),
	# PRIMARY TR: noise-only TR (action N(0,0.05) on throttle and steering; tau0 N(1000,50), omega0 N(2600,100))
	# plus a random steering-rate limit, LogUniform(0.05, 0.5) normalized units per 0.1 s control step, one draw per trial
	'tr_rate': (TRIALS, {**REL, 'steer_rate': ('loguniform', 0.05, 0.5)}),
	# secondary: noise-only TR (action and powertrain noise, no rate limit)
	'paper_tr': (TRIALS, REL),
	# secondary: sensitivity of the primary TR to the range and the shape of the rate-limit distribution
	'tr_rate_half': (TRIALS, {**REL, 'steer_rate': ('loguniform', 0.025, 0.25)}),
	'tr_rate_double': (TRIALS, {**REL, 'steer_rate': ('loguniform', 0.1, 1.0)}),
	'tr_rate_uniform': (TRIALS, {**REL, 'steer_rate': ('uniform', 0.05, 0.5)}),
	# secondary: the other actuator family (random delay of 0-3 control steps) with the noise-only TR
	'tr_delay': (TRIALS, {**REL, 'delay': ('choice', [0, 1, 2, 3])}),
	# secondary deterministic baselines (not TR): the release Chrono loop's fixed rate limit 0.1;
	# the fixed rate limit with the 1 s start hold and the HMMWV wheelbase (3.378 m)
	'det_rate_0.1': (1, {'steer_rate': ('const', 0.1)}),
	'det_calibrated': (1, {'wheelbase': ('const', 3.378), 'steer_rate': ('const', 0.1), 'hold': ('const', 10)}),
	}