13. Non-contact/Contact
Transitions
12
Intuition Method Experiments
Endow the robotic system with a controller such that:
(I) Ensuring stable contact,
(II) At a desired location,
(III) Leaving on the surface at a desired point
Task objectives
➢The impact happens only once
➢The contact and departure points
Mirrazavi Salehian, S. S. and Billard, A. (2018) A Dynamical System Based Approach for Controlling Robotic Manipulators During Non-contact/Contact Transitions. RA-L
𝑥 𝑐
𝑥 𝑙
Contact surface
➢ Non-penetrable, passive and convex.
26. Non-contact/Contact
Transitions
Intuition Method Experiments
𝑥 𝑐
18
𝑥 𝑙 𝑞1 𝑞2
Legend
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
Task objectives
➢The impact happens only once
➢The contact and stop/departure points
27. Non-contact/Contact
Transitions
Intuition Method Experiments
𝑥 𝑐
18
𝑥 𝑙 𝑞1 𝑞2
Legend
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
Task objectives
➢The impact happens only once
➢The contact and stop/departure points
Elastic impact
28. Non-contact/Contact
Transitions
Intuition Method Experiments
𝑥 𝑐
18
𝑥 𝑙 𝑞1 𝑞2
Legend
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
Task objectives
➢The impact happens only once
➢The contact and stop/departure points
Task objectives
➢The impact happens only once: 𝑞1
𝑇
ሶ𝑥 𝑡∗ = 0
Elastic impact
29. Non-contact/Contact
Transitions
Intuition Method Experiments
𝑥 𝑐
18
𝑥 𝑙 𝑞1 𝑞2
Legend
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
Task objectives
➢The impact happens only once
➢The contact and stop/departure points
Task objectives
➢The impact happens only once: 𝑞1
𝑇
ሶ𝑥 𝑡∗ = 0
Elastic impact
ሶ𝑥 = 𝑀 𝑥 𝑓 𝑥
30. Non-contact/Contact
Transitions
Intuition Method Experiments
𝑥 𝑐
18
𝑥 𝑙 𝑞1 𝑞2
Legend
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
Task objectives
➢The impact happens only once
➢The contact and stop/departure points
Task objectives
➢The impact happens only once: 𝑞1
𝑇
ሶ𝑥 𝑡∗ = 0
Elastic impact
ሶ𝑥 = 𝑀 𝑥 𝑓 𝑥
We must control for the robot’s
velocity at contact.
31. Non-contact/Contact
Transitions
Intuition Method Experiments
𝑥 𝑐
18
𝑥 𝑙 𝑞1 𝑞2
Legend
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
Task objectives
➢The impact happens only once
➢The contact and stop/departure points
Task objectives
➢The impact happens only once: 𝑞1
𝑇
ሶ𝑥 𝑡∗ = 0
Elastic impact
ሷ𝑥 = 𝑀 𝑥, ሶ𝑥 𝑓 𝑥, ሶ𝑥, 𝑡
We must control for the robot’s
velocity at contact.
32. Non-contact/Contact
Transitions
Intuition Method Experiments
31
ሷ𝑥 = 𝑓 𝑥, ሶ𝑥ሶ𝑥 = 𝑓 𝑥
➢ Input: Position
➢ Output: Velocity
➢ Control variable is Position
➢ Demonstrations must cover position
space.
➢ Input: Position and Velocity
➢ Output: Acceleration
➢ Control variable is Velocity and
indirectly Position.
➢ Demonstrations must cover position and
velocity space.
33. Non-contact/Contact
Transitions
Intuition Method Experiments
𝑥 𝑐
32
𝑀 𝑥, ሶ𝑥 = 𝑄Λ𝑄−1
𝑄 = 𝑞1 ⋯ 𝑞 𝑑
𝑇
ሷ𝑥 = 𝑀 𝑥, ሶ𝑥 𝑓 𝑥, ሶ𝑥, 𝑡
Locally activeDirectional Modulation
𝑥 𝑙 𝑞1 𝑞2
𝑥 𝑐
Mirrazavi Salehian, S. S. and Billard, A. (2018) A Dynamical System Based Approach for Controlling Robotic Manipulators
During Non-contact/Contact Transitions. RA-L
Legend
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
34. Legend
Non-contact/Contact
Transitions
Intuition Method Experiments
𝑞1
33
𝑀 𝑥, ሶ𝑥 = 𝑄Λ𝑄−1
𝑄 = 𝑞1 ⋯ 𝑞 𝑑
𝑇
ሷ𝑥 = 𝑀 𝑥, ሶ𝑥 𝑓 𝑥, ሶ𝑥, 𝑡
Locally activeDirectional Modulation
𝑞2
𝑥 𝑐
𝑥 𝑙
Mirrazavi Salehian, S. S. and Billard, A. (2018) A Dynamical System Based Approach for Controlling Robotic Manipulators
During Non-contact/Contact Transitions. RA-L
33
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
35. Non-contact/Contact
Transitions
Intuition Method Experiments
34
ሷ𝑥 = 𝑀 𝑥, ሶ𝑥 𝑓 𝑥, ሶ𝑥, 𝑡
Locally active
𝑥 𝑐
𝑥 𝑙 𝑞1 𝑞2
Free motion region:
𝜌 ≤ Γ 𝑥
Transition/contact regions:
0 ≤ Γ 𝑥 < 𝜌
Λ = I → M = 𝑄𝑄−1
ሷ𝑥 = 𝑓 𝑥, ሶ𝑥, 𝑡 Λ =
𝜆11 ⋯ 𝜆1𝑑
⋮ ⋱ ⋮
𝜆 𝑑1 ⋯ 𝜆 𝑑𝑑
Legend
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
00:15.510
𝑀
𝐼
𝜌 Γ 𝑥
Λ
Mirrazavi Salehian, S. S. and Billard, A. (2018) A Dynamical System Based Approach for Controlling Robotic Manipulators
During Non-contact/Contact Transitions. RA-L
36. Legend
Non-contact/Contact
Transitions
Intuition Method Experiments
35
𝑀 𝑥, ሶ𝑥 = 𝑄Λ𝑄−1
𝑄 = 𝑞1 ⋯ 𝑞 𝑑
𝑇
ሷ𝑥 = 𝑀 𝑥, ሶ𝑥 𝑓 𝑥, ሶ𝑥, 𝑡
Locally activeDirectional Modulation
𝑥 𝑐
𝑥 𝑙
Mirrazavi Salehian, S. S. and Billard, A. (2018) A Dynamical System Based Approach for Controlling Robotic Manipulators
During Non-contact/Contact Transitions. RA-L
35
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
Λ =
𝜆11 ⋯ 𝜆1𝑑
⋮ ⋱ ⋮
𝜆 𝑑1 ⋯ 𝜆 𝑑𝑑
lim
𝑡→∞
𝑞𝑖
𝑇
𝑥 − 𝑞𝑖
𝑇
𝑥 𝑐 = 0
∀𝑖 ∈ 2 … 𝑑
lim
𝑡→∞
𝑞1
𝑇
𝑥 = 0
𝑞1
𝑇
ሶ𝑥 𝑡∗
= 0
37. Legend
Non-contact/Contact
Transitions
Intuition Method Experiments
29
ሷ𝑥 = 𝑀 𝑥, ሶ𝑥 𝑓 𝑥, ሶ𝑥, 𝑡
Locally activeDirectional Modulation
𝑥 𝑐
𝑥 𝑙
Mirrazavi Salehian, S. S. and Billard, A. (2018) A Dynamical System Based Approach for Controlling Robotic Manipulators
During Non-contact/Contact Transitions. RA-L
36
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
lim
𝑡→∞
𝑞𝑖
𝑇
𝑥 − 𝑞𝑖
𝑇
𝑥 𝑙 = 0
∀𝑖 ∈ 2 … 𝑑
lim
𝑡→∞
𝑞1
𝑇
𝑥 = 0
𝑞1
𝑇
ሶ𝑥 𝑡∗
= 0
𝑀 𝑥, ሶ𝑥 = 𝑄Λ𝑄−1
𝑄 = 𝑞1 ⋯ 𝑞 𝑑
𝑇
Λ =
𝜆11 ⋯ 𝜆1𝑑
⋮ ⋱ ⋮
𝜆 𝑑1 ⋯ 𝜆 𝑑𝑑
38. Legend
Non-contact/Contact
Transitions
Intuition Method Experiments
29
ሷ𝑥 = 𝑀 𝑥, ሶ𝑥 𝑓 𝑥, ሶ𝑥, 𝑡
Locally activeDirectional Modulation
𝑥 𝑐
𝑥 𝑙
Mirrazavi Salehian, S. S. and Billard, A. (2018) A Dynamical System Based Approach for Controlling Robotic Manipulators
During Non-contact/Contact Transitions. RA-L
37
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
lim
𝑡→∞
𝑞𝑖
𝑇
𝑥 − 𝑞𝑖
𝑇
𝑥 𝑙 = 0
∀𝑖 ∈ 2 … 𝑑
lim
𝑡→∞
𝑞1
𝑇
𝑥 = 0
𝑞1
𝑇
ሶ𝑥 𝑡∗
= 0
𝑀 𝑥, ሶ𝑥 = 𝑄Λ𝑄−1
𝑄 = 𝑞1 ⋯ 𝑞 𝑑
𝑇
Λ =
𝜆11 ⋯ 𝜆1𝑑
⋮ ⋱ ⋮
𝜆 𝑑1 ⋯ 𝜆 𝑑𝑑
39. Legend
Non-contact/Contact
Transitions
Intuition Method Experiments
29
ሷ𝑥 = 𝑀 𝑥, ሶ𝑥 𝑓 𝑥, ሶ𝑥, 𝑡
Locally activeDirectional Modulation
𝑥 𝑐
𝑥 𝑙
Mirrazavi Salehian, S. S. and Billard, A. (2018) A Dynamical System Based Approach for Controlling Robotic Manipulators
During Non-contact/Contact Transitions. RA-L
38
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
lim
𝑡→∞
𝑞𝑖
𝑇
𝑥 − 𝑞𝑖
𝑇
𝑥 𝑙 = 0
∀𝑖 ∈ 2 … 𝑑
lim
𝑡→∞
𝑞1
𝑇
𝑥 = 0
𝑞1
𝑇
ሶ𝑥 𝑡∗
= 0
𝑀 𝑥, ሶ𝑥 = 𝑄Λ𝑄−1
𝑄 = 𝑞1 ⋯ 𝑞 𝑑
𝑇
Λ =
𝜆11 ⋯ 𝜆1𝑑
⋮ ⋱ ⋮
𝜆 𝑑1 ⋯ 𝜆 𝑑𝑑
40. Legend
Non-contact/Contact
Transitions
Intuition Method Experiments
29
ሷ𝑥 = 𝑀 𝑥, ሶ𝑥 𝑓 𝑥, ሶ𝑥, 𝑡
Locally activeDirectional Modulation
𝑥 𝑐
𝑥 𝑙
Mirrazavi Salehian, S. S. and Billard, A. (2018) A Dynamical System Based Approach for Controlling Robotic Manipulators
During Non-contact/Contact Transitions. RA-L
39
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
lim
𝑡→∞
𝑞𝑖
𝑇
𝑥 − 𝑞𝑖
𝑇
𝑥 𝑙 = 0
∀𝑖 ∈ 2 … 𝑑
lim
𝑡→∞
𝑞1
𝑇
𝑥 = 0
𝑞1
𝑇
ሶ𝑥 𝑡∗
= 0
Free motion region:
𝜌 ≤ Γ 𝑥
Λ = I M = I
ሷ𝑥 = 𝑓 𝑥, ሶ𝑥, 𝑡
𝑀 𝑥, ሶ𝑥 = 𝑄Λ𝑄−1
𝑄 = 𝑞1 ⋯ 𝑞 𝑑
𝑇
Λ =
𝜆11 ⋯ 𝜆1𝑑
⋮ ⋱ ⋮
𝜆 𝑑1 ⋯ 𝜆 𝑑𝑑
41. Legend
Non-contact/Contact
Transitions
Intuition Method Experiments
29
ሷ𝑥 = 𝑀 𝑥, ሶ𝑥 𝑓 𝑥, ሶ𝑥, 𝑡
Locally activeDirectional Modulation
𝑥 𝑐
𝑥 𝑙
Mirrazavi Salehian, S. S. and Billard, A. (2018) A Dynamical System Based Approach for Controlling Robotic Manipulators
During Non-contact/Contact Transitions. RA-L
40
𝑥 Robot’s state
Γ(x) The surface contact
𝑥 𝑐
Desired contact point
𝑥 𝑙
Desired departure point
lim
𝑡→∞
𝑞𝑖
𝑇
𝑥 − 𝑞𝑖
𝑇
𝑥 𝑙 = 0
∀𝑖 ∈ 2 … 𝑑
lim
𝑡→∞
𝑞1
𝑇
𝑥 = 0
𝑞1
𝑇
ሶ𝑥 𝑡∗
= 0
Free motion region:
𝜌 ≤ Γ 𝑥
Λ = I M = I
ሷ𝑥 = 𝑓 𝑥, ሶ𝑥, 𝑡
𝑀 𝑥, ሶ𝑥 = 𝑄Λ𝑄−1
𝑄 = 𝑞1 ⋯ 𝑞 𝑑
𝑇
Λ =
𝜆11 ⋯ 𝜆1𝑑
⋮ ⋱ ⋮
𝜆 𝑑1 ⋯ 𝜆 𝑑𝑑