- 12
- 6 098
JunYoung Kim
South Korea
Приєднався 7 тра 2022
Mecanum Wheeled-Mobile Robot: Reactive Motion to External Force Feedback
Based on the compliance control, the mobile robot reactively moves in response to external force feedback.
00:00 Prototype1
00:23 Prototype2
00:44 Prototype2(demo)
00:54 end
00:00 Prototype1
00:23 Prototype2
00:44 Prototype2(demo)
00:54 end
Переглядів: 217
Відео
Humanoid Robot G1, ZMP-based Walking
Переглядів 1,5 тис.2 місяці тому
Offline ZMP-based Walking using Weighted Whole-body Control. Swing Leg Trajectory is computed using Cubic Spline Polynomial. The simulation was conducted with fixed single and double support phases with varied step lengths.
Humanoid Robot G1, Stepping Boxes using Weighted WBC
Переглядів 5102 місяці тому
Stepping box scenario using weighted whole-body control approach. The desired end-effector position was pre-defined.
Humanoid Robot G1, Whole-Body Motion at Single Stance
Переглядів 8302 місяці тому
Unitree G1, model-based Whole-Body Motion Control at Single Stance (End-effector, CoM, pitch, roll, yaw motion). Weighted WBC Approach with the following costs and constraints: - Costs: Centroidal Dynamics Task, Joint Acceleration Task, Regularization Task, End-effector Task - Equality: Floating-base Dynamics, Contact Constraints - Inequality: Friction Cone Constraints - Dynamics Library: Custo...
Reactive Walking on a Random Height Field
Переглядів 3842 місяці тому
Controlled based on Online Reactive Stepping Planner and Whole-Body Impulse Control - DCM-based online walking with an optimization approach. - Bipedal robot has total 6DoF, 3DoF for each leg. - The simulation was conducted using MuJoCo, based on our custom articulated rigid body dynamics library.
DCM-based Reactive Walking of Pointfoot Bipedal Robot
Переглядів 4034 місяці тому
Controlled based on Online Reactive Stepping Planner and Whole-Body Impulse Control - DCM-based walking, next footstep position, and step duration are computed online with an optimization approach. - Bipedal robot has total 6DoF, 3DoF for each leg. - The simulation was conducted using MuJoCo, with the dynamics library based on our custom articulated rigid body dynamics library. - A total of 4 Q...
Pointfoot Walking in Sagittal Plane, based on Capture Point
Переглядів 1234 місяці тому
Capture Point-based walking in Sagittal Plane ( Walking on the spot and Forward Walking ) - Template point-foot model has 2DoF for each leg - Control based on joint Trajectory planning and kinematic controller - Recovery from the external pushes
Humanoid Robot G1, model-based Whole-Body Motion Control
Переглядів 1,6 тис.4 місяці тому
Humanoid G1, model-based Whole-Body Motion Control using custom Kinematics and Dynamics (ARBMLlib) library. Weighted WBC Approach with the following costs and constraints: - Costs: Centroidal Dynamics Task, Joint Acceleration Task, Regularization Task - Equality: Floating-base Dynamics, Contact Constraints - Inequality: Friction Cone Constraints - QP Solver: qpOASES(c )
CoM shifting in the z-axis, based on Task-Prioritized WBC
Переглядів 1158 місяців тому
CoM shifting in the z-axis, based on Task-Prioritized WBC in 4 point contact
CoM shifting in the x-axis, based on Task-Prioritized WBC
Переглядів 1018 місяців тому
CoM shifting in the x-axis, with fixed hand position task, based on Task-Prioritized WBC in 4 point contact
Humanoid Robot Localization
Переглядів 213Рік тому
A vision-based localization system for the humanoid robot platform. Integrated a particle filter with a CNN-based object detection model to achieve precise localization. Also, improved data association process by applying Hungarian algorithm to optimally match the landmark with detected keypoints. This resulted in precise localization and enabled to earn RoboCup 2022 2nd Place.
0:24 What are you doing step-robot?
Thank you very much. I noticed in your recent work that you've combined ZMP and WBC approaches, which is an excellent idea. I have two questions that I would like to ask: 1. Are you using the ZMP method to offline plan a desired trajectory as a reference, and then applying WBC for online optimization? 2. As far as I know, the trajectories planned by ZMP are typically position-based, including the center of mass (CoM) and foot trajectories. How did you design the WBC problem in this case? Does WBC output torque commands? How do you convert position information to torque?
Thank you for your question! 1. As mentioned in the description of the walking video, the desired trajectory was planned in offline manner. 2. To explain briefly, the decision variables of the WBC include joint accelerations and contact forces. From these, we can compute the joint torques using inverse dynamics. The following references might help you design a WBC: arxiv.org/abs/1410.7284 ieeexplore.ieee.org/document/5648837 dspace.mit.edu/bitstream/handle/1721.1/128291/1201260987-MIT.pdf?sequence=1&isAllowed=y
@@junyoungkim_robotics Okay, thanks. I'll have a try.
I am interested in looking at the code, can you provide me your git hub link.
Can you help me to control this model. Please
What kind of help do you need?
Which sensors you used and algorithm control (PID, Scurve motion or sliding mode .....)? Thank you very much
Hello, may I ask what type of actuator you are using from mujoco?
I’ve been using the model from the following link: github.com/unitreerobotics/unitree_mujoco/tree/main/unitree_robots/g1, which utilizes torque commands as input.
캬
excellent work! would love to see if you could share training video of how to do this kind of robot simulation in mujoco
Wonderful
Awesome!
Awesome
오 잘되네
감사합니다ㅎㅎ
What is ARBMLlib and I can't find an introduction to this library.
It’s a custom rigid body dynamics library, similar to RBDL or Pinocchio. Unfortunately, ours is not opend to the public.
what simulator are you using?
This was simulated at MuJoCo.
@@junyoungkim_robotics do you think Drake is capable to visualize trajectories as shown in your Mujoco ?
@@bilynbk I’m not certain about that since I’ve never tried Drake before.