def reset_world(self, world, epoch):
world.origin = np.array([world.size / 2, world.size / 2])
# agent colors
for i, agent in enumerate(world.agents):
if agent.adversary:
# agent.color = world.predator_colors[i]
agent.color = np.array([0.85, 0.35, 0.35])
else:
agent.color = np.array([0.35, 0.85, 0.35])
# properties for landmarks
for i, landmark in enumerate(world.landmarks):
landmark.color = np.array([0.25, 0.25, 0.25])
# generate predators in random circle of random radius with random angles
redraw = True
while redraw:
# draw location for prey
prey_pt = world.origin + np.random.normal(0.0, 0.0001, size=2)
# prey_pt = np.array([0., 0.])
# draw predator locations
if world.init_pos_curriculum:
world.pred_init_distance = world.pred_init_distance_end - (
world.pred_init_distance_end -
world.pred_init_distance_start) * max(
(world.decay - epoch) / world.decay, 0.0)
init_pts = [
world.origin +
np.random.normal(0.0, world.pred_init_distance, size=2)
for _ in range(self.n_preds)
]
else:
init_pts = [
np.random.uniform(0.0, world.size, size=2)
for _ in range(self.n_preds)
]
# init_pts = [np.array([5., 3.]), np.array([2., 4.73205081]), np.array([1.99999999, 1.2679492 ])]
# init_pts = [np.array([7., 5.]), np.array([4., 6.73205081]), np.array([3.99999999, 3.2679492 ])]
# init_pts = [np.array([1., 3.]), np.array([3., 5.]), np.array([5., 3])]
# ensure predators not initialized on top of prey
redraw = overlaps(prey_pt, init_pts, world.size, threshold=0.5)
# set initial states
init_pts.append(prey_pt)
for i, agent in enumerate(world.agents):
agent.active = True
agent.captured = False
# agents can move beyond confines of camera image --> need to adjust coords accordingly
agent.state.coords = init_pts[i]
agent.state.p_pos = agent.state.coords % world.size
agent.state.p_vel = np.zeros(world.dim_p)
agent.state.theta = 0.0
agent.state.c = np.zeros(world.dim_c)
def reset_world(self, world):
world.origin = np.array([world.size/2, world.size/2])
print('world size = {}'.format(world.size))
print('pred vel = {}'.format(world.agents[0].max_speed))
print('prey vel = {}'.format(world.agents[-1].max_speed))
print('tax = {}'.format(world.tax))
print('action penalty = {}'.format(world.action_penalty))
print('capture reward = {}'.format(world.cap_reward))
# agent color
for i, agent in enumerate(world.agents):
if agent.adversary:
agent.color = world.predator_colors[i]
else:
agent.color = np.array([0.35, 0.85, 0.35])
# properties for landmarks
for i, landmark in enumerate(world.landmarks):
landmark.color = np.array([0.25, 0.25, 0.25])
# generate predators in random circle of random radius with random angles
redraw = True
while redraw:
# draw location for prey
prey_pt = world.origin + np.random.normal(0.0, 0.0001, size=2)
# draw predator locations
init_pts = [np.random.uniform(0.0, world.size, size=2) for _ in range(self.n_preds)]
# init_pts = [np.array([1.5, 1.5]) + np.random.normal(0.0, 0.01, size=2), np.array([1.5, 1.5])+np.random.normal(0.0, 0.01, size=2), np.array([1.5, 1.5])]
# ensure predators not initialized on top of prey
redraw = overlaps(prey_pt, init_pts, world.size, threshold=0.5)
# set initial states
init_pts.append(prey_pt)
for i, agent in enumerate(world.agents):
agent.active = True
agent.captured = False
# agents can move beyond confines of camera image --> need to adjust coords accordingly
agent.state.coords = init_pts[i]
agent.state.p_pos = agent.state.coords % world.size
agent.state.p_vel = np.zeros(world.dim_p)
agent.state.theta = 0.0
agent.state.c = np.zeros(world.dim_c)