How to change environment by adding dynamic and static obstacles?

[RuxueYan]

I've changed the environment using examples in the usage of intelligent-robot-simulator, but where can I implement this environment in gym_env and policy_train_process? Thank you.

Basically, how should we add all the varied environments like dynamic obstacles and different shapes of obstacles from 'Intelligent robot simulator' to this project? i.e. which file do we need to add or edit, or the reward function...

[hanruihua]

Hi, thanks for your interest on my work. To change the env in the training process, you can modify the file _trainworld.yaml directly in the policytrain folder or using your own yaml file by changing the argument --worldpath in the train_process.py.

Most of the environment can be set by the yaml file, such as the different size of obstacles. But for the dynamic obstacles, the function _obs_cirsstep is necessary to assignment velocity of obstacles. All these modifications can be transformed in the _stepir function in the mrnav.py

[RuxueYan]

Hi, thank you for your reply. I've successfully added dynamic obstacles to the environment under gym_test_world.yaml and gy_env_test.py by adding obs_cirs_stpe() in step_ir, but when I tried to run train_process the dynamic obstacles seem static, do I need to edit ppo.training_loop? And for static obstacles, it seems that the environment doesn't reset if robots collide with static obstacles, do I need to write a new VO reward function to fix it? Thank you very much.

[hanruihua]

Hi, I have added a file _dynamic_obstest.yaml to set the world with dynamic obstacles and added _obs_cirsstep() in step_ir in the latest version.

It works well. You can change the --world_path to _dynamic_obstest.yaml to use without other changes. The rvo reward function calculates these obstacles too.

[RuxueYan]

Hi, thank you so much for your help. But I noticed that dynamic obstacles on plots of train_process and pre_trained_model still seem static (obstacles don't move at all for an epoch), even though robots did avoid those obstacles in pre_trained_model. Is this normal? Thank you.

[hanruihua]

Hi, the normal animation with dynamic obstacles should be as follows, which is generated with the latest version by policy_test_pre_train.py

[RuxueYan]

Hi ruihua, It works well now. Thank you for your help, I will close the issue.