def set_target(self, target):
self.target = target
self.env.gs[target] = grid_spec.REWARD
self.q_values = q_iteration.q_iteration(env=self.env,
num_itrs=32,
discount=0.99)
self.env.gs[target] = grid_spec.EMPTY
def _new_target(self, start, target):
# print('Computing waypoints from %s to %s' % (start, target))
start = self.gridify_state(start)
start_idx = self.env.gs.xy_to_idx(start)
target = self.gridify_state(target)
target_idx = self.env.gs.xy_to_idx(target)
self._waypoint_idx = 0
self.env.gs[target] = grid_spec.REWARD
q_values = q_iteration.q_iteration(env=self.env,
num_itrs=50,
discount=0.99)
# compute waypoints by performing a rollout in the grid
max_ts = 100
s = start_idx
waypoints = []
for i in range(max_ts):
a = np.argmax(q_values[s])
new_s, reward = self.env.step_stateless(s, a)
waypoint = self.env.gs.idx_to_xy(new_s)
if new_s != target_idx:
waypoint = waypoint - np.random.uniform(size=(2, )) * 0.2
waypoints.append(waypoint)
s = new_s
if new_s == target_idx:
break
self.env.gs[target] = grid_spec.EMPTY
self._waypoints = waypoints
self._waypoint_prev_loc = start
self._target = target