lgsvl/gym-lgsvl

OpenAI Gym Environment for SVL Simulator.

Reinforcement Learning with OpenAI Gym

See detailed instructions and tutorial at: https://www.svlsimulator.com/docs/third-party-integration/openai-gym

gym-lgsvl's Issues

Bug: possible typo in collision callback registration

I did not try out anything, but i am pretty sure that the on_collision callback is not correctly registered due to a typo at line 83.

ModuleNotFoundError: No module named 'spinup'

The installation instruction don't mention to install spinningup. I just commented it out in random_agent.py, since it is only used for logging, to get the agent running. But I think either should it be mentioned in the instructions or replaced by some more "light weight" logging mechanism.

gym.error.UnregisteredEnv: No registered env with id: gym_lgsvl:lgsvl-v0

When i run ./random_agent.py, it throws an error:

gym.error.UnregisteredEnv: No registered env with id: gym_lgsvl:lgsvl-v0

Here are the traceback:

Traceback (most recent call last):
File "/home/bob/anaconda3/lib/python3.5/site-packages/gym/envs/registration.py", line 143, in spec
return self.env_specs[id]
KeyError: 'gym_lgsvl:lgsvl-v0'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "./random_agent.py", line 6, in
env = gym.make('gym_lgsvl:lgsvl-v0')
File "/home/bob/anaconda3/lib/python3.5/site-packages/gym/envs/registration.py", line 167, in make
return registry.make(id)
File "/home/bob/anaconda3/lib/python3.5/site-packages/gym/envs/registration.py", line 118, in make
spec = self.spec(id)
File "/home/bob/anaconda3/lib/python3.5/site-packages/gym/envs/registration.py", line 153, in spec
raise error.UnregisteredEnv('No registered env with id: {}'.format(id))
gym.error.UnregisteredEnv: No registered env with id: gym_lgsvl:lgsvl-v0

Running this on Ubuntu 16.04.

Rewardfunction Lanefollowing Waypointtargeting

Dear LGSVL Team,

Currently we are developing an lanefollowing reward function. It will reward our agent when it stays in its lane. For this we target the waypoints on the lanesegment "segment.TargetLocalPosition". We calculate the distance between the ego position and the nearest waypoint on the lane with regard to the agent in world coordinates. But we have a problem evaluating the correct waypoint which is beeing targeted in the UnitySimulator. Because the transform.position of the waypoints in the inspector of the UnitySimulator shows only LocalPosition.

We are thinking about spawning an object at the position of the currentTarget for visualization in game scene. Is there a better and easier solution you would consider evaluating the correct target of the agent?

Best Regards,
qsengshu

OpenAI /Gym : No registered env with id: BanditTenArmedGaussian-v0

Hi,

I have windows 10 (64x), Jupyter Notebook and Python 3.

I downloaded gym_bandits from here

git clone https://github.com/JKCooper2/gym-bandits.git
cd gym-bandits
pip install -e .

My code :

import gym
import gym_bandits
import numpy as np
import math
import random
env = gym.make("BanditTenArmedGaussian-v0")

And I have a error I don't understand why and this package is not widely use so I can't find any other report to help me to fix it.
That's why I'm here.

---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
c:\windows\system32\gym\gym\envs\registration.py in spec(self, path)
    120         try:
--> 121             return self.env_specs[id]
    122         except KeyError:

KeyError: 'BanditTenArmedGaussian-v0'

During handling of the above exception, another exception occurred:

UnregisteredEnv                           Traceback (most recent call last)
<ipython-input-15-ec95f73807de> in <module>
      4 import math
      5 import random
----> 6 env = gym.make("BanditTenArmedGaussian-v0")

c:\windows\system32\gym\gym\envs\registration.py in make(id, **kwargs)
    143 
    144 def make(id, **kwargs):
--> 145     return registry.make(id, **kwargs)
    146 
    147 def spec(id):

c:\windows\system32\gym\gym\envs\registration.py in make(self, path, **kwargs)
     87         else:
     88             logger.info('Making new env: %s', path)
---> 89         spec = self.spec(path)
     90         env = spec.make(**kwargs)
     91         # We used to have people override _reset/_step rather than

c:\windows\system32\gym\gym\envs\registration.py in spec(self, path)
    129                 raise error.DeprecatedEnv('Env {} not found (valid versions include {})'.format(id, matching_envs))
    130             else:
--> 131                 raise error.UnregisteredEnv('No registered env with id: {}'.format(id))
    132 
    133     def register(self, id, **kwargs):

UnregisteredEnv: No registered env with id: BanditTenArmedGaussian-v0

AttributeError: 'LgsvlEnv' object has no attribute 'camera'

Hi,
I have tried to execute gym-lgsvl.
When I have execute the following command
python -m baselines.run --alg=a2c --env=gym_lgsvl:lgsvl-v0 --num_timesteps=1e5
I got an Attribute Error
AttributeError: 'LgsvlEnv' object has no attribute 'camera'
(check the screenshot)

Please help me to solve this

Thanks in Advance
Malathi K

Target directory should be created

Right now, random_agent.py expects a directory ~/gym-lgsvl to be existing to save the tmp.png to.
I put my gym in a different directory and consequently got a confusing error message until I changed the target directory.

cv2.error: OpenCV(4.1.0) /io/opencv/modules/imgproc/src/resize.cpp:3718: error: (-215:Assertion failed) !ssize.empty() in function 'resize'

I'd suggest that either the target directory should be created by the script or, even better, the image should be written to /tmp.

cant add additional sensors

hey,
when im loading an agent through _setup_ego function, which I provide a configuration id as the below example:
def _setup_ego(self, name = "30387e1f-4d7f-4b6e-98aa-d721a16c4e75", spawn_index = 0, random_spawn = False):

while the name argument is the configuration ID set on web UI, which loaded with several sensors.
but when I am printing all the available sensors, not all of them are exist in the jason file, alltough the simulator itself can show me all the sensors

I'm printing them using the below code:

def _setup_ego(self, name = "30387e1f-4d7f-4b6e-98aa-d721a16c4e75", spawn_index = 0, random_spawn = False):

    state = lgsvl.AgentState()
    if (random_spawn):
      state.transform = self.spawns[random.randint(0, len(self.spawns) - 1)]
    else:
      state.transform = self.spawns[spawn_index]
    
    self.ego = self.env.add_agent(name, lgsvl.AgentType.EGO, state)
    self.vehicles[self.ego] = "EGO"
    self._occupied.append(state.transform.position)
    self.sensors = self.ego.get_sensors()
    for s in self.sensors:
      print("(_setup_ego) sensor: " ,s.name)
      if (s.name == "Color Camera Sensor"):
        self.camera = s
        break
    self.ego_position = state.transform.position

for example,
I've added the lane sensor, and it does not print it as a sensor, while on the simulator UI i can see the lane sensor

Setup Python API and running on Server

The page linked to set up the python API is missing. Can u give an effective one?
Another question is that can we use this lgsvl environment on server which doesn't have any GUI?

Thanks.