From 291fd6fab232910b580b50bf871930d3a735381c Mon Sep 17 00:00:00 2001
From: Andrea Prestia <67275846+andreh1111@users.noreply.github.com>
Date: Tue, 1 Sep 2026 15:02:06 -0400
Subject: [PATCH 1/3] fix doubled render call

---
 robohive/envs/env_base.py | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/robohive/envs/env_base.py b/robohive/envs/env_base.py
index bc3cc13f..115538c3 100644
--- a/robohive/envs/env_base.py
+++ b/robohive/envs/env_base.py
@@ -293,6 +293,8 @@ def step(self, a, **kwargs):
                                         step_duration=self.dt,
                                         realTimeSim=self.mujoco_render_frames,
                                         render_cbk=self.mj_render if self.mujoco_render_frames else None)
+        # robot.step() above already rendered this tick when mujoco_render_frames is True
+        self._rendered_this_tick = self.mujoco_render_frames
         return self.forward(**kwargs)
 
     @implement_for("gym", None, "0.24")
@@ -317,9 +319,10 @@ def _forward(self, **kwargs):
         Returns current obs(t), rwd(t), done(t), info(t)
         """
 
-        # render the scene
-        if self.mujoco_render_frames:
+        # render the scene (skipped if step() already rendered this tick)
+        if self.mujoco_render_frames and not getattr(self, '_rendered_this_tick', False):
             self.mj_render()
+        self._rendered_this_tick = False
 
         # observation
         obs = self.get_obs(**kwargs)

From 351c6cebad5c055882cee433233f9eb8b8a7d98b Mon Sep 17 00:00:00 2001
From: Andrea Prestia <67275846+andreh1111@users.noreply.github.com>
Date: Tue, 1 Sep 2026 15:49:31 -0400
Subject: [PATCH 2/3] address vk comment

---
 robohive/envs/env_base.py | 28 +++++++++++++++-------------
 1 file changed, 15 insertions(+), 13 deletions(-)

diff --git a/robohive/envs/env_base.py b/robohive/envs/env_base.py
index 115538c3..d1fd9a5d 100644
--- a/robohive/envs/env_base.py
+++ b/robohive/envs/env_base.py
@@ -92,6 +92,7 @@ def _setup(self,
         robot_cls = kwargs.pop('robot_cls', Robot)
         self.robot = robot_cls(mj_sim=self.sim,
                            random_generator=self.np_random,
+                           env_dt=self.sim.model.opt.timestep * frame_skip,
                            **kwargs)
 
         #resolve action space
@@ -293,36 +294,37 @@ def step(self, a, **kwargs):
                                         step_duration=self.dt,
                                         realTimeSim=self.mujoco_render_frames,
                                         render_cbk=self.mj_render if self.mujoco_render_frames else None)
-        # robot.step() above already rendered this tick when mujoco_render_frames is True
-        self._rendered_this_tick = self.mujoco_render_frames
-        return self.forward(**kwargs)
+        # robot.step() above already rendered this tick (render_cbk in hardware mode,
+        # sim.advance(render=True) in sim mode) -- forward()'s own render would
+        # otherwise duplicate it, uncompensated by robot.step()'s pacing sleep.
+        return self.forward(render=False, **kwargs)
 
     @implement_for("gym", None, "0.24")
-    def forward(self, **kwargs):
-        return self._forward(**kwargs)
+    def forward(self, render=True, **kwargs):
+        return self._forward(render=render, **kwargs)
 
     @implement_for("gym", "0.24", None)
-    def forward(self, **kwargs):
-        obs, reward, done, info = self._forward(**kwargs)
+    def forward(self, render=True, **kwargs):
+        obs, reward, done, info = self._forward(render=render, **kwargs)
         terminal = done
         return obs, reward, terminal, False, info
 
     @implement_for("gymnasium")
-    def forward(self, **kwargs):
-        obs, reward, done, info = self._forward(**kwargs)
+    def forward(self, render=True, **kwargs):
+        obs, reward, done, info = self._forward(render=render, **kwargs)
         terminal = done
         return obs, reward, terminal, False, info
 
-    def _forward(self, **kwargs):
+    def _forward(self, render=True, **kwargs):
         """
         Forward propagate env to recover env details
         Returns current obs(t), rwd(t), done(t), info(t)
         """
 
-        # render the scene (skipped if step() already rendered this tick)
-        if self.mujoco_render_frames and not getattr(self, '_rendered_this_tick', False):
+        # render the scene -- render=False from step(), which already rendered via
+        # robot.step(); standalone callers (e.g. after reset()) keep the default.
+        if self.mujoco_render_frames and render:
             self.mj_render()
-        self._rendered_this_tick = False
 
         # observation
         obs = self.get_obs(**kwargs)

From 4479f6e21a176f83aa371a3b11030de37bd5cdf5 Mon Sep 17 00:00:00 2001
From: Andrea Prestia <67275846+andreh1111@users.noreply.github.com>
Date: Tue, 1 Sep 2026 15:52:13 -0400
Subject: [PATCH 3/3] revert mistakenly committed update

---
 robohive/envs/env_base.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/robohive/envs/env_base.py b/robohive/envs/env_base.py
index d1fd9a5d..96d3b7b7 100644
--- a/robohive/envs/env_base.py
+++ b/robohive/envs/env_base.py
@@ -92,7 +92,6 @@ def _setup(self,
         robot_cls = kwargs.pop('robot_cls', Robot)
         self.robot = robot_cls(mj_sim=self.sim,
                            random_generator=self.np_random,
-                           env_dt=self.sim.model.opt.timestep * frame_skip,
                            **kwargs)
 
         #resolve action space