Tri Wahyu Guntara

Kitchen enivronment observation is 60 instead of 30

1

Hi, I am wondering whether including goal as an input is intentional. The SPIRL paper says that the kitchen environment has 30 states but running `python3 spirl/train.py --path=spirl/configs/skill_prior_learning/kitchen/hierarchical_cl --val_data_size=160` will...

I am trying to implement meta-learning for RL that needs environment that can be applied with certain damages. I am trying to implement locked joint on the agent e.g. ant...

Wrong conversion from CARLA coordinate system to ROS coordinate system?

4

Hi, I noticed that in [carla_rotation_to_RPY](https://github.com/carla-simulator/ros-bridge/blob/master/carla_common/src/carla_common/transforms.py#L81) function, **yaw and pitch are inverted**. If CARLA uses left-hand coordinate system (with z-axis up) and ROS uses right-hand coordinate system (with z-axis up),...

Changing the callback mechanism of `AdversarialTrainer` such that we can insert `sb3.EvalCallback`. See #607.

How to properly inserts sb3 `EvalCallback` for `AdversarialTraining`?

5

Seems like there is no easy way to insert `EvalCallback` to `AdversarialTraining`, no? `AdversarialTraining.train_gen()` populates `callback` argument of `sb3.BaseAlgorithm.learn()` with `self.gen_callback` which is either `None` or `reward_wrapper.RewardVecEnvWrapper()`. As an additional...

Minor bug: agent is using cuda:0 device no matter what rl_device arg is

2

### Problem - `ase.learning.common_agent.CommonAgent` inherits `rl_games.common.a2c_common.A2CBase` which stores all tensors to `self.ppo_device`. - `self.ppo_device` is set by getting `device` key from `config`. If there is no `device` key, it is...

Difference between state, observation, and amp_observation

1

Hi, I would like to ask the difference between states, observations, and amp_observations. My understanding is that the state space is not defined for the humanoid task and just the...

I run the code as is `python main.py` with `Python3.6` and `tensorflow1.4.0` but the result is not even close. Do you know why? The plot looks like that perhaps because...

create_graph parameter is False hence first-order MAML?

1

Hi, I noticed that when using `torch.autograd.grad`, the implementation does not pass `create_graph` parameter which by default is `False`. See [L#86](https://github.com/dragen1860/MAML-Pytorch/blob/master/meta.py#L86) and [L#116](https://github.com/dragen1860/MAML-Pytorch/blob/master/meta.py#L116) of the `Meta` class. To compute higher-order...

Inquiry: vertex corresponds to deterministic policy and representation of reward

1

Hello Benjamin Eysenbach, I have two questions about the paper. I would really appreciate it if you have the time to explain it. 1. In section 5.1, it is stated...