From 95d2dc949cef5342cb6db732f236d102bc6ba6fb Mon Sep 17 00:00:00 2001
From: Dipam Chakraborty <dipam@aicrowd.com>
Date: Thu, 11 Nov 2021 12:46:16 +0530
Subject: [PATCH] round 2 env with action repeats every 30ms
---
.gitignore | 3 +-
.../deepracer_gym/envs/deepracer_gym_env.py | 32 ++++++++++++++++++-
deepracer-gym/random_actions_example.py | 8 +++--
deepracer-gym/start_deepracer_docker.sh | 2 +-
4 files changed, 40 insertions(+), 5 deletions(-)
diff --git a/.gitignore b/.gitignore
index 8686387..42968a8 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,3 +1,4 @@
__pycache__
deepracer-gym/deepracer_gym.egg-info/
-playground/
\ No newline at end of file
+playground/
+deepracer-gym/dump.py
diff --git a/deepracer-gym/deepracer_gym/envs/deepracer_gym_env.py b/deepracer-gym/deepracer_gym/envs/deepracer_gym_env.py
index e497b6c..72dfb58 100644
--- a/deepracer-gym/deepracer_gym/envs/deepracer_gym_env.py
+++ b/deepracer-gym/deepracer_gym/envs/deepracer_gym_env.py
@@ -1,17 +1,47 @@
import numpy as np
import gym
from deepracer_gym.zmq_client import DeepracerEnvHelper
-
+import time
+import warnings
class DeepracerGymEnv(gym.Env):
def __init__(self):
self.action_space = gym.spaces.Discrete(5)
self.deepracer_helper = DeepracerEnvHelper()
+ self.last_step_time = None
+ self.max_step_time = 0.03 # seconds
def reset(self):
observation = self.deepracer_helper.env_reset()
return observation
def step(self, action):
+
+ if self.last_step_time is not None:
+ time_delta = time.time() - self.last_step_time - self.max_step_time
+ else:
+ time_delta = -1
+ done = False
+ n_repeats = 0
+
+ # This is to emulate async nature of the real world track
+ # If action is not returned within required time limit, the same action would be repeated
+ while (not done) and time_delta > 0:
+ time_delta -= self.max_step_time
+ observation, reward, done, info = self._step_sim(self.last_action)
+ n_repeats += 1
+
+ if n_repeats > 0:
+ warn_msg = f"Action was repeated {n_repeats} times, try to reduce model step time to {self.max_step_time} seconds"
+ warnings.warn(warn_msg)
+
+ if not done:
+ observation, reward, done, info = self._step_sim(action)
+
+ self.last_action = action
+ self.last_step_time = time.time()
+ return observation, reward, done, info
+
+ def _step_sim(self, action):
rl_coach_obs = self.deepracer_helper.send_act_rcv_obs(action)
observation, reward, done, info = self.deepracer_helper.unpack_rl_coach_obs(rl_coach_obs)
return observation, reward, done, info
diff --git a/deepracer-gym/random_actions_example.py b/deepracer-gym/random_actions_example.py
index c496ee6..61de439 100644
--- a/deepracer-gym/random_actions_example.py
+++ b/deepracer-gym/random_actions_example.py
@@ -2,6 +2,7 @@ import gym
import numpy as np
import deepracer_gym
+import time
env = gym.make('deepracer_gym:deepracer-v0')
@@ -13,8 +14,11 @@ steps_completed = 0
episodes_completed = 0
total_reward = 0
-for _ in range(500):
+for i in range(500):
observation, reward, done, info = env.step(np.random.randint(5))
+ # Example to show the delay in model inference
+ if (i-5) % 5 == 0:
+ time.sleep(0.1)
steps_completed += 1
total_reward += reward
@@ -23,4 +27,4 @@ for _ in range(500):
episodes_completed += 1
print("Episodes Completed:", episodes_completed, "Steps:", steps_completed, "Reward", total_reward)
steps_completed = 0
- total_reward = 0
\ No newline at end of file
+ total_reward = 0
diff --git a/deepracer-gym/start_deepracer_docker.sh b/deepracer-gym/start_deepracer_docker.sh
index cb04278..dd2f02a 100644
--- a/deepracer-gym/start_deepracer_docker.sh
+++ b/deepracer-gym/start_deepracer_docker.sh
@@ -1 +1 @@
-docker run -it --rm --name=deepracer -p 8888:8888 -p 5000:5000 --cpus="3" --memory="6g" aicrowd/base-images:deepracer_round1_release /bin/bash
+docker run -it --rm --name=deepracer -p 8888:8888 -p 5000:5000 --cpus="3" --memory="6g" aicrowd/base-images:deepracer_round2_release /bin/bash
--
GitLab