diff --git a/addons/python/cfg/pendulum_sarsa.yaml b/addons/python/cfg/pendulum_sarsa.yaml
new file mode 100644
index 00000000..8550bb24
--- /dev/null
+++ b/addons/python/cfg/pendulum_sarsa.yaml
@@ -0,0 +1,67 @@
+experiment: 
+  type: experiment/online_learning
+  runs: 1
+  trials: 1000
+  steps: 0
+  rate: 0
+  test_interval: 10
+  environment: 
+    type: environment/modeled
+    model: 
+      type: model/dynamical
+      control_step: 0.03
+      integration_steps: 5
+      dynamics: 
+        type: dynamics/pendulum
+    task: 
+      type: task/pendulum/swingup
+      timeout: 2.99
+      randomization: 0
+  agent: 
+    type: agent/td
+    policy: 
+      type: mapping/policy/discrete/value/q
+      discretizer: 
+        type: discretizer/uniform
+        min: experiment/environment/task/action_min
+        max: experiment/environment/task/action_max
+        steps: [3]
+      projector: 
+        type: projector/tile_coding
+        tilings: 16
+        memory: 8388608
+        safe: 0
+        resolution: [0.31415, 3.1415, 3]
+        wrapping: [6.283, 0, 0]
+      representation: 
+        type: representation/parameterized/linear
+        init_min: [0]
+        init_max: [1]
+        memory: experiment/agent/policy/projector/memory
+        outputs: 1
+        output_min: []
+        output_max: []
+      sampler: 
+        type: sampler/epsilon_greedy
+        rand_max: 0
+        epsilon: 0.05
+    predictor: 
+      type: predictor/critic/sarsa
+      alpha: 0.2
+      gamma: 0.97
+      lambda: 0.65
+      projector: experiment/agent/policy/projector
+      representation: experiment/agent/policy/representation
+      trace: 
+        type: trace/enumerated/replacing
+  test_agent: 
+    type: agent/fixed
+    policy: 
+      type: mapping/policy/discrete/value/q
+      discretizer: experiment/agent/policy/discretizer
+      projector: experiment/agent/policy/projector
+      representation: experiment/agent/policy/representation
+      sampler: 
+        type: sampler/greedy
+        rand_max: 0
+  save_every: never
diff --git a/addons/python/share/experiment_example.py b/addons/python/share/experiment_example.py
new file mode 100755
index 00000000..0cf52030
--- /dev/null
+++ b/addons/python/share/experiment_example.py
@@ -0,0 +1,14 @@
+#!/usr/bin/python3
+#
+# GRL experiment example
+
+import gym
+import grlpy
+
+# Create GRL agent
+conf = grlpy.Configurator("../cfg/pendulum_sarsa.yaml")
+inst = conf.instantiate()
+exp = grlpy.Experiment(inst["experiment"])
+exp.run()
+exp.reset()
+exp.run()
diff --git a/addons/python/src/grlpy.cpp b/addons/python/src/grlpy.cpp
index 49114da2..8a02edec 100644
--- a/addons/python/src/grlpy.cpp
+++ b/addons/python/src/grlpy.cpp
@@ -94,6 +94,9 @@ PYBIND11_MODULE(grlpy, m) {
         mycfg.set(item.first.attr("__str__")().cast<std::string>(), item.second.attr("__str__")().cast<std::string>());
   
         conf.reconfigure(mycfg);
+      })
+    .def("reset", [](Configurable &conf) {
+        conf.reset();
       });
 
   // Experiment
diff --git a/externals/pybind11/include/pybind11/detail/class.h b/externals/pybind11/include/pybind11/detail/class.h
index 7a5dd013..060c0a80 100644
--- a/externals/pybind11/include/pybind11/detail/class.h
+++ b/externals/pybind11/include/pybind11/detail/class.h
@@ -222,7 +222,9 @@ inline bool deregister_instance_impl(void *ptr, instance *self) {
     auto &registered_instances = get_internals().registered_instances;
     auto range = registered_instances.equal_range(ptr);
     for (auto it = range.first; it != range.second; ++it) {
-        if (Py_TYPE(self) == Py_TYPE(it->second)) {
+        // #2252 / 2b6b98e
+        //if (Py_TYPE(self) == Py_TYPE(it->second)) {
+        if (self == it->second) {
             registered_instances.erase(it);
             return true;
         }