vincentyu68 / policy_transfer Goto Github PK

The paper mentions that the PPO was run for 500 iterations, with each iteration sampling 40,000 steps from the environment.
Does it mean that run_ppo.py has to be run with a batch size of 40000 and max_step of 2e7 ?

After setting simulator parameters using set_simulator_parameter, get_simulator_parameter does not return the expected values.
This might be due to a logical error in the code in the file parameter_managers.py, class walker2dParamManager.
The set_simulator_parameter

for jid in range(7, 13):<br/>
            if jid in self.controllable_param:
                damp = x[cur_id] * (self.damping_range[1] - self.damping_range[0]) +self.damping_range[0]
                self.simulator.robot_skeleton.joints[jid - 5].set_damping_coefficient(0, damp)

updates joints[2] to joints[7]
Whereas the get_simulator_parameter

for jid in range(3, 9):
            cur_damp = self.simulator.robot_skeleton.joints[jid].damping_coefficient(0)

returns joints[3] to joints[8].

Could you please rectify this?

Also, what are the 8 controllable simulator parameters did you control to obtain the results in the paper?