Fixed bug when buffer not reinitialized on reset

README.md

@@ -570,6 +570,8 @@ However, note that doing this would imply using a longer action buffer.
 
 #### Bonus 3: Pro tips
 
+##### a) Elasticity
+
 The time-step's maximum elasticity defines the tolerance of your environment in terms of time-wise precision.
 It is set in the configuration dictionary as the `"time_step_timeout_factor"` entry.
 This can be any value `> 0.0`.
@@ -590,6 +592,14 @@ You may want this to be as tight as possible.
 In such situation, keep in mind that inference must end before the end of this next time-step, since the computed action is to be applied there.
 Otherwise, your time-steps will timeout.
 
+##### b) Reset
+
+In `rtgym`, the default action is sent when `reset()` is called.
+This is to maintain the real-time flow of time-steps during reset transitions.
+
+It may happen that you prefer to repeat the previous action instead, for instance because it is hard in your application to implement a no-op action.
+
+To achieve this behavior, you can simply replace the `default_action` attribute of your environment with the action that you want being sent, right before calling `reset()`.
 
 ---
 

rtgym/envs/real_time_env.py

@@ -476,6 +476,8 @@ def reset(self):
         self.current_step = 0
         if self.reset_act_buf:
             self.init_action_buffer()
+        else:
+            self.act_buf.append(self.default_action)
         elt = self.interface.reset()
         if self.act_in_obs:
             elt = elt + list(self.act_buf)

setup.py

@@ -7,14 +7,14 @@
 
 setup(name='rtgym',
       packages=[package for package in find_packages()],
-      version='0.5',
+      version='0.6',
       license='MIT',
       description='Easily implement custom OpenAI Gym environments for real-time applications',
       long_description=long_description,
       long_description_content_type="text/markdown",
       author='Yann Bouteiller',
       url='https://github.com/yannbouteiller/rtgym',
-      download_url='https://github.com/yannbouteiller/rtgym/archive/refs/tags/v0.5.tar.gz',
+      download_url='https://github.com/yannbouteiller/rtgym/archive/refs/tags/v0.6.tar.gz',
       keywords=['gym', 'real', 'time', 'custom', 'environment', 'reinforcement', 'learning', 'random', 'delays'],
       install_requires=['gym', 'numpy'],
       classifiers=[