Metrics

Metrics are automatically computed and available in info[0]['metrics'] on the final step of an episode.

Classical MAPF Metrics (on_target='nothing')

Metric	Key	Description	Range
CSR	'CSR'	1.0 if all agents are on goals simultaneously	{0, 1}
ISR	'ISR'	Fraction of agents on goals at episode end	[0, 1]
EpLength	'EpLength'	Total episode steps	[1, max_steps]
SumOfCosts	'SumOfCosts'	Total steps across all agents	[num_agents, ...]
Makespan	'Makespan'	Steps until last agent finishes	[1, max_steps]

Lifelong Metrics (on_target='restart')

Metric	Key	Description
LifeLongAverageThroughput	'avg_throughput'	(Total goals reached) / max_steps

Finish Metrics (on_target='finish')

Metric	Key	Description	Range
CSR	'CSR'	1.0 if ALL agents reached goals	{0, 1}
ISR	'ISR'	Fraction of agents that reached goals	[0, 1]
EpLength	'EpLength'	Average steps to reach goal per agent	[1, max_steps]
SumOfCosts	'SumOfCosts'	Total steps across all agents	[num_agents, ...]
Makespan	'Makespan'	Steps until last agent finishes	[1, max_steps]

Accessing Metrics

from pogema import pogema_v0, GridConfig

env = pogema_v0(GridConfig(num_agents=4, size=8, seed=42))
obs, info = env.reset()

while True:
    obs, reward, terminated, truncated, info = env.step(env.sample_actions())
    if all(terminated) or all(truncated):
        break

metrics = info[0]['metrics']
print(f"CSR: {metrics['CSR']}, ISR: {metrics['ISR']}")