Presentation is loading. Please wait.

Presentation is loading. Please wait.

Introduction to Manipulation

Similar presentations


Presentation on theme: "Introduction to Manipulation"— Presentation transcript:

1 Introduction to Manipulation
Lecture 15-1 Introduction to Manipulation

2 Introduction to Manipulation
Manipulation Definition - 물리적 접촉이나 힘과 모멘트에 영향을 미침으로 Object들을 움직이거나 재배치 하는 과정 (원문 : refer to the process of moving or rearranging objects in the environment by establishing physical contact and exerting forces and moments) Problem Formulation == Walking parameterrize - m:mass, I:Inertia xy plane orientation theta F1, F2, F3 목표 - Object의 초기 자세 에서( 𝑞 𝑜𝑏𝑗𝑒𝑐𝑡 , 𝑞 𝑟𝑜𝑏𝑜𝑡 ) 목표 위치(?)로 움직이기( 𝑞𝑑 𝑜𝑏𝑗𝑒𝑐𝑡 ) Walking Problem과의 차이점 - Kinematics와 Dynamics가 더 Decomposition 됨 - More Contact and richer models of contact(non-convex geometry) - More focus on uncertainty(relatively less on stability)

3 Kinematics Planning A Standard Decomposition
Step 1 : examine object and figure out where/ how to grasp it “grasp synthesis” Step 2 : plan collision free kinematics trajectory to grasp location or pre grasp Step 3 : execute the grasp Step 4 : plan collision free path for robot + object Step 5 : release grasp

4 Kinematics Planning Kinematics Planning(via trajectory optimization)
𝑞 =𝑢 : direct control velocity of the object Robot Configuration : 𝑥 ℎ𝑎𝑛𝑑 =𝑓(𝑞) : the position od the hand or each fingers<- forward kinematics Kinematics Constraints : 𝑥 𝑓𝑖𝑛𝑔𝑒𝑟𝑠 𝑡 𝑝𝑖𝑐𝑘𝑢𝑝 =𝑔𝑟𝑎𝑠𝑝 𝑝𝑜𝑖𝑛𝑡 : nonlinear constraint Avoid Collision->Joint limit->solver and trajectory optimization with constraint

5 Grasp Synthesis : Form Closure
Grasp Synthesis/Grasp Feedback From Closure : for all object motions 𝑞 𝑜𝑏𝑗𝑒𝑐𝑡≠0 then at least one contact constraint is violated Notation - Normal Vector ni, Contact Points Pi, Contact Point in body frame ri

6 Grasp Synthesis : Form Closure
Single Contact point 𝑛𝑖 𝑇 𝑃𝑖 ≥0 Pi=F(qobject, ri) 𝑃𝑖 = 𝜕𝑓 𝜕𝑞 𝑞 𝑜𝑏𝑗𝑒𝑐𝑡=𝐽𝑖(𝑞𝑜𝑏𝑗𝑒𝑐𝑡, 𝑟𝑖) 𝑞 𝑜𝑏𝑗𝑒𝑐𝑡 Goal->to pick ri 𝑛𝑖 𝑇 𝐽(𝑞,𝑟𝑖) 𝑞 ≥0 -> linear eq. in 𝑞 -> for all 𝑞 , must violate at least one constraint A 𝑞 ≥0 => 𝑞 =0 - linear program

7 Grasp Synthesis: Force Closure
Lecture 15-4 Grasp Synthesis: Force Closure

8 Grasp Synthesis: Force Closure
Force와 Torque의 관계식(자코비안) 과 마찰력 조건을 이용하여 로봇 손이 인가해야 할 힘을 계산한다. Wrench란?? F or T?? 최적화를 통해 minimum force를 구할 수 있다. Grasp feedback을 이용하여 외란에도 원하는 동작을 수행할 수 있도록 한다.

9 Grasp Synthesis: Force Closure
Contact position 설정이 중요하다.

10 Grasp Synthesis: Force Closure
Underactuated hand (trend)  why? 1 motor, 4-bar linkage + spring Atlas hand DRC-HUBO hand

11 Manipulation with Dynamic Constraints
Lecture 15-5 Manipulation with Dynamic Constraints

12 Manipulation with Dynamic Constraints
The little humanoid of the FIRA soccer competition. 만약에 물체가 로봇보다 무거우면? 무거운 물체를 들면 로봇의 다이나믹스도 변한다.

13 Manipulation with Dynamic Constraints
𝑞 𝑟𝑜𝑏𝑜𝑡 , 𝑞 𝑜𝑏𝑗𝑒𝑐𝑡 는 항상 알고 있고, 𝑞 𝑟𝑜𝑏𝑜𝑡 0 , 𝑞 𝑜𝑏𝑗𝑒𝑐𝑡 (0)이 주어졌을 때 trajectory optimization을 통해 입력 u를 찾을 수 있다. Dynamics constraints와 𝑞 𝑜𝑏𝑗𝑒𝑐𝑡 𝑡 𝑓𝑖𝑛𝑎𝑙 =𝑔𝑜𝑎𝑙 을 만족해야 한다.

14 Manipulation with Dynamic Constraints
이렇게 복잡한 contact model을 전에 배웠던 hybrid event로 쉽게 계산할 수 있을까? LCP(Linear Complenentary Problem) solver

15 Manipulation with Dynamic Constraints
충돌할 때의 힘을 구할 수 있다면 시뮬레이션은 쉽게 만들어진다. 아래의 조건들을 만족하는 힘 F를 Optimization을 통해 구할 수 있다. 만약 마찰력이 선형화되었다면 complementary constraint도 역시 선형 시스템이기 때문에 Linear complementary problem(LCP)라 한다. (If I make a linear approximation of my friction cone, it’s LCP.) 매 시간 LCP를 푼다. F는 0  Impulse로 발생. : not good. “time-stepping method” (using time-averaged force)

16 Manipulation with Dynamic Constraints
𝑥 𝑛+1 = 𝑓 𝑑𝑡 𝑥 𝑛 ,𝑢 𝑛 ,𝐹 𝑛 SNOPT (Sparse Nonlinear Optimizer)  라이브러리!!


Download ppt "Introduction to Manipulation"

Similar presentations


Ads by Google