Skip to content

behavior_env_demo

main(random_selection=False, headless=False, short_exec=False)

Generates a BEHAVIOR Task environment from a pre-defined configuration file.

It steps the environment 100 times with random actions sampled from the action space, using the Gym interface, resetting it 10 times.

Source code in examples/environments/behavior_env_demo.py 
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
def main(random_selection=False, headless=False, short_exec=False):
    """
    Generates a BEHAVIOR Task environment from a pre-defined configuration file.

    It steps the environment 100 times with random actions sampled from the action space,
    using the Gym interface, resetting it 10 times.
    """
    logging.info("*" * 80 + "\nDescription:" + main.__doc__ + "*" * 80)

    # Ask the user whether they want online object sampling or not
    sampling_options = {
        False: "Use a pre-sampled cached BEHAVIOR activity scene",
        True: "Sample the BEHAVIOR activity in an online fashion",
    }
    should_sample = choose_from_options(options=sampling_options, name="online object sampling", random_selection=random_selection)

    # Load the pre-selected configuration and set the online_sampling flag
    config_filename = os.path.join(og.example_config_path, "fetch_behavior.yaml")
    cfg = yaml.load(open(config_filename, "r"), Loader=yaml.FullLoader)
    cfg["task"]["online_object_sampling"] = should_sample

    # If we're online sampling, make sure global contacts are enabled so we can accurately detect kinematic changes
    if should_sample:
        gm.ENABLE_GLOBAL_CONTACT_REPORTING = True

    # Load the environment
    env = og.Environment(configs=cfg)

    # Allow user to move camera more easily
    og.sim.enable_viewer_camera_teleoperation()

    # Run a simple loop and reset periodically
    max_iterations = 10 if not short_exec else 1
    for j in range(max_iterations):
        logging.info("Resetting environment")
        env.reset()
        for i in range(100):
            action = env.action_space.sample()
            state, reward, done, info = env.step(action)
            if done:
                logging.info("Episode finished after {} timesteps".format(i + 1))
                break

    # Always close the environment at the end
    env.close()