Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

GAZEBO simulator #6

Open
Geonhee-LEE opened this issue Aug 25, 2020 · 8 comments
Open

GAZEBO simulator #6

Geonhee-LEE opened this issue Aug 25, 2020 · 8 comments
Assignees

Comments

@Geonhee-LEE
Copy link
Owner

GAZEBO simulator를 기반으로 테스트 진행

@CzJaewan
Copy link
Collaborator

GAZEBO simulator, Global Plan(A*) Following with RL

model : servingbot
git : https://github.com/CzJaewan/servingbot

global_plan(A*) and RL test code
git : https://github.com/CzJaewan/rl_avoidance_gazebo.git

@CzJaewan
Copy link
Collaborator

CzJaewan commented Sep 1, 2020

ppo used look a head : https://www.youtube.com/watch?v=7nEgkgscECQ
ppo : https://www.youtube.com/watch?v=1_WY7lGVc-o&feature=youtu.be, https://www.youtube.com/watch?v=7Y1Mcp_e418

@Geonhee-LEE
Copy link
Owner Author

Geonhee-LEE commented Sep 1, 2020

굿굿 장애물(gazebo에서 박스)이 추가했을때의 결과도 궁금 + actor라는 움직이는 물체도 gazebo에서 사용가능http://gazebosim.org/tutorials?tut=actor&cat=build_robot
human.zip 파일

@Geonhee-LEE Geonhee-LEE reopened this Sep 1, 2020
@Geonhee-LEE
Copy link
Owner Author

@CzJaewan GAZEBO에서 돌릴 수 있는 multi model launch방법 안내부탁드려요~ 다음주부터 진행 예정

@CzJaewan
Copy link
Collaborator

@Geonhee-LEE
Copy link
Owner Author

Geonhee-LEE commented Sep 15, 2020

현재(2020.09.15)까지 진행사항:

  • rl_collision_avoidance를 기반으로 학습 완료.
    • 시간에 대한 페널티 추가.
    • wander 추가
  • 가지보 시뮬레이션에서 look ahead를 목표지점으로 하여 물체 회피 진행
    • actor(사람) 추가하여 테스트 진행: 동작확인, 그러나 완벽하지 않으며 아슬하게 앞쪽으로 회피하는 경우 발생
    • 타겟(look ahead)이 지속적으로 변경되는데 이것이 영향을 주진 않는 지 확인 필요
    • AMCL 로컬라이제이션이 정상적으로 작동되지 않는 것처럼 보임, 위치 추정이 잘 안되거나 갑자기 snap(위치 점프)이 발생
    • 하나의 런치 파일로 통합필요, repo 정리.
    • prediction python 파일은 로스로 실행하고 토픽을 주고 받을 수 있도록 구성해야함.

진행 중인 사항

  • multi-agent 실행
  • 기존에는 라이더만 검출, 라이더의 위치를 변경하여 로봇이 검출되도록 변경
  • 라이더 스캔 데이터의 시간 차이가 작아, 물체의 동적 특징을 잡지 못할 수가 있음 -> 스텝을 주어 테스트 진행 예정.

진행 예정 사항

  • Hybrid control
    • 로컬 플래너를 플러그인을 통해 사용해야하는데 c++만 가능, python으로 실행하는 것을 어떻게 병합할 지 고려.
  • 실제 로봇 포팅

Repository owner deleted a comment from CzJaewan Sep 15, 2020
Repository owner deleted a comment from CzJaewan Sep 15, 2020
Repository owner deleted a comment from CzJaewan Sep 15, 2020
@CzJaewan
Copy link
Collaborator

CzJaewan commented Sep 16, 2020

우선순위

1. Gazebo wander env test

  • observation study : lidar scan data의 시간 간격 찾기.
  • reward setting : lidar distance reward 효율성 찾기
  • DWA와 비교

2. 실제 servingbot wander world test

  • 실제 실험

3. multi agent env test (syscon bot)

  • wander world 학습 policy와 비교

4. 실제 syscon bot multi agent world test

5. hybrid control

  • mpc와 순수 rl tracking과 비교

SAC, TD3 등의 알고리즘 변경

LSTM 도입 network

  • lidar의 시간별 data 분류

@CzJaewan
Copy link
Collaborator

ppo world sub data
object_state_topic : [pose : /odom.pose], [speed : /odom.twist] -> [pose : /amcl_pose.pose], [speed : /odom.twist]

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

2 participants