Sorry but I tried lots of different algorithms(D4PG,DDPG,SAC) with solid implementations( they worked & converged in other envs).

This environment has either problem with reward assignment or something else. I trained each one for 8 Million steps(10 000 episodes) none of them came close to beating getting more than -2.0 reward.

While videos looks promising I think its something environment details fundamentally broken.

I can run the PPO.py, and can see the process of training. But after training for about a couple of hours, my computer(the GPU is GeForce 1060) would breakdown.

So, I just want to test the model you have trained well. Could you share it ? Thanks!

Hi,

Looks like great work, and helps advance an idea I had and wanted to expand upon!

Just slightly confused how I'm editing the state variables to change VEL_STATE and also Discrete vs Continuous. Any help would be appreciated!

Thanks for all the work!

Anthony

Running your PPO project, I got a series of very black frames. It's hard to find the rocket.

So I decided to test the environment by newing a python file and added these code:

          import gym
          import time

          env = gym.make('RocketLander-v0')
          env.reset()
          for t in range(500):
               env.render()
               env.step(env.action_space.sample())
               time.sleep(0.05)

I had the same problem ! How should I solve this problem ? Please ! ! !

When is try to make env with gym I get:-

pybox2d/library/Box2D/Box2D.py", line 158, in _init_kwargs
% (self.__class__.__name__, key, ex))
AssertionError: Failed on kwargs for b2PolygonShape.vertices: count >= 3

flake8 testing of https://github.com/EmbersArc/gym on Python 2.7.14

$ flake8 . --count --select=E901,E999,F821,F822,F823 --show-source --statistics

./envs/board_game/go.py:108:60: F821 undefined name 'BLACK'
    curr_state = GoState(pachi_py.CreateBoard(board_size), BLACK)
                                                           ^
./envs/board_game/go.py:111:53: F821 undefined name 'BLACK'
        a = (black_policy_fn if curr_state.color == BLACK else white_policy_fn)(curr_state, prev_state, prev_action)
                                                    ^
2     F821 undefined name 'BLACK'

How to fix it 👍
Traceback (most recent call last):
File "E:\HOC\nam4\spacex\gym-master\envs\box2d\rocket_lander.py", line 3, in
import Box2D
ModuleNotFoundError: No module named 'Box2D'