def compute_reward(self, rl_actions, **kwargs):
"""See class definition."""
if self.env_params.evaluate:
return - rewards.min_delay_unscaled(self)
else:
return (- rewards.min_delay_unscaled(self) +
rewards.penalize_standstill(self, gain=0.2))
def compute_reward(self, rl_actions, **kwargs):
"""See class definition."""
r = 0
if rl_actions is not None:
r = -rewards.boolean_action_penalty(rl_actions >= 0.5, gain=2)
if self.env_params.evaluate:
r += -rewards.min_delay_unscaled(self)
#print(f"Reward computed: {r}, rl_actions: {rl_actions}")
else:
r += (-rewards.min_delay_unscaled(self) +
rewards.penalize_standstill(self, gain=0.2))
print(f"Reward computed: {r}, rl_actions: {rl_actions}")
return r
def compute_reward(self, rl_actions, **kwargs):
"""See class definition."""
if self.env_params.evaluate:
return -rewards.min_delay_unscaled(self)
else:
"""
print('delay penalty:', -rewards.min_delay_unscaled(self))
print('standstill penalty:', rewards.penalize_standstill(self, gain=0.2))
print('action penalty:', -rewards.boolean_action_penalty(rl_actions >= 0.5, gain=0.2))
"""
#return (- rewards.min_delay_unscaled(self) +
# rewards.penalize_standstill(self, gain=0.2))
return (
-rewards.min_delay_unscaled(self) -
rewards.boolean_action_penalty(rl_actions >= 0.5, gain=0.01) -
rewards.waiting_penalty(self, gain=0.01))
def compute_reward(self, rl_actions, **kwargs):
"""See class definition."""
if self.test:
return 0
return - rewards.min_delay_unscaled(self) \
- rewards.boolean_action_penalty(rl_actions >= 0.5, gain=1.0)
def compute_reward(self, rl_actions, **kwargs):
"""See class definition."""
if rl_actions is None:
return {}
if self.env_params.evaluate:
rew = -rewards.min_delay_unscaled(self)
else:
rew = -rewards.min_delay_unscaled(self) \
+ rewards.penalize_standstill(self, gain=0.2)
# each agent receives reward normalized by number of lights
rew /= self.num_traffic_lights
rews = {}
for rl_id in rl_actions.keys():
rews[rl_id] = rew
return rews
def compute_reward(self, rl_actions, **kwargs):
"""See class definition."""
if self.env_params.evaluate:
return -rewards.min_delay_unscaled(self)
else:
return rewards.desired_velocity(self, fail=kwargs["fail"])
def compute_reward(self, state, rl_actions, **kwargs):
if self.env_params.evaluate:
return rewards.min_delay_unscaled(self)
else:
return rewards.desired_velocity(self, fail=kwargs["fail"])