search

duckietown / gym-duckietown

Self-driving car simulator for the Duckietown universe
http://duckietown.org
Other
51 stars 19 forks source link

Issue with self._drivable_pos(f_pos) #53

Closed bhairavmehta95 closed 6 years ago

bhairavmehta95 commented 6 years ago

Found by Adam when testing ./manual_control --map_name 4way_adam

When we hit the 4 way intersection in the middle of the map, self._drivable_pos(f_pos) turns to false (I checked all of the conditions in _valid_pose() return ( self._drivable_pos(self.cur_pos) and # True self._drivable_pos(l_pos) and # True self._drivable_pos(r_pos) and # True self._drivable_pos(f_pos) and # Turns to False not self._collision() # True aka hasn't collided with anything. )

Investigating what the issue is, but if you could also take a look @maximecb that'd be great

bhairavmehta95 commented 6 years ago

Issue is that the 4way tile is not considered to be drivable for some reason.

Happens in both 4way_adam and 4way.

Reason: Line 378 in simplesim_env.py doesn't process the 4way tile correctly. I can fix it (add an elif tile == '4way'), but I'm not sure what to put for the orient and angle attributes.

maximecb commented 6 years ago

I would just edit the map files and put /N for now.

maximecb commented 6 years ago

Maybe we should also have a regression test map for this, regress_4way. Could be a map with just a 4way tile, to force the duckie to spawn there, and we do a step with 0 velocity, check that the episode doesn't terminate.

maximecb commented 6 years ago

I added a little regression test that triggers this bug. Testing is important y'all!

https://github.com/duckietown/gym-duckietown/blob/8de0bba8e3236cef320b31d34c5bb3215ec61402/gym_duckietown/maps/regress_4way_drivable.yaml

maximecb commented 6 years ago

Sidenote: maps with names beginning with regress_ will now be loaded in the tests but not in the multi map training environment.