Generality and Simple Hands

Transcription

1 Generality and Simple Hands with Grasp Invariance Matthew T. Mason and Alberto Rodriguez February 18, 2011

2 Outline 1. Overview of Robotics Institute (Mason) 2. Simple hands project (Mason) 3. Grasp invariance (Rodriguez)

3 The Robotics Institute Founded 30 years ago as a research institute. The first Robotics Department, for 20 years. but still more of a research institute. About 500 researchers: 15% faculty (2/3 on soft money) 25% graduate students, 45% technical staff, Sponsored research budget: $55M 15% visiting faculty and postdocs.

4 70,000 Robotics Institute Annual Expenditures by Funding Source ($000's) 60,000 50,000 Contributions, Fellowships, Endowment University $000's 40,000 30,000 20,000 Other Non-Federal Industry Other Federal NSF 10,000 NASA DOD Fiscal Year

5 RI research foci Manufacturing, Logistics & Scheduling. Mobile Robotics and Field Robotics. Vision, Perception in general. Learning. Human Robot Interaction. Medical Robotics. Quality of Life. Legged Locomotion, Manipulation. Snakes. Computer Graphics, Animation, Humanoids.

6 The Urban Challenge 3 November, 2007 Autonomous driving in urban environment. 11 vehicles qualified for the final event. 6 vehicles finished. 3 vehicles required no interventions. None were perfect; all were great. Carnegie Mellon finished in ~ 4 hours and won by ~20 minutes.

7

8 What type of robot hand do we want? This? Or this? The Kraft Viper It depends!

9 Can a robot hand be both simple and general? Assume: Few actuators, e.g. one; Few sensors; Simple mechanisms; Small, light, inexpensive. Could it... Pick parts from a bin? Operate scissors? Open a door? Fold origami? The Kraft Viper

10 Another one

11 Six simple but Sony SMART Cell not general hands.

12 The case for complexity. In examples, complexity correlates with generality. Grasping involves conforming hand shape to given object shape. More freedoms implies greater variety of shapes. In-hand manipulation of a rigid body the straightforward way requires nine actuators. For haptic shape sensing, more sensors and more freedoms implies more data. Design constraints have consequences.

13 Second, the case for simple hands...

14 In favor of simplicity... da Vinci Surgical System Kanazawa University

15 Some simple but general hands. da Vinci surgery (and origami) human with prosthetic hook hardware pickup tool chopsticks roulette croupier underwater teleoperated grippers

16 The case for simplicity. Humans, animals, teleoperators can do a lot with simple hands; Practical issues robustness, cost, weight,... Scientific benefits. The gold standard for generality is a human with tools. Humans are adept with anthropomorphic and simple effectors.

17 Scenario Bin picking with a Simple Hand. Simple mechanism and control. Bin-pick highlighters. Indirectly: Dimensions of task generality. Design and Manufacturing of Simple Hands.

18

19 Approach I. Simple mechanism

20 Approach II. Simple control Let the fingers fall where they may vs. Put the fingers on the right place

21 Approach II. Simple control

22 Approach III. Offline Learning Grasp Classification In-hand Localization

23 Approach IV. Reiteration III. Offline Learning II. Simple control I. Simple mechanism

24 Prototype I 3 fingers. Gear transmission. Torsional springs.

25 Prototype II 4 fingers. Linkage transmission. Elastic link in the linkage. Fully observable.

26 Grasp Signature FINGER 4 FINGER 3 FINGER 2 FINGER 1 MOTOR

27 Experimental Setting Industrial manipulator. Preprogrammed grasp motion. State machine commands: Robot Gripper Vision system Logger 200 trials with each gripper.

28 Grasp Trials

29 Grasp Recognition PCA compression. SVM classifier. %

30 Grasp Recognition Tradeoff precision-recall Precision Recall

31 In-Hand Localization Locally Weighted Regression Average Error ( ) Recall

32 In-Hand Localization

33 Discussion/Future Work Bin picking addressable with Simple Hand. Mechanism: Non-interfering fingers. Palm and finger shape: V-shape potential fields. Variable stiffness. Control: Blind policy (this paper) Early abort (current work) Learn policy from a small parametrized set.

34 Grasp Invariance

35

36

37 Structure Part I - The Problem Part II - The Solution Part III - The Applications

38 Part I - Problem

39 Finger Form

40 Some finger forms... Simple hands Closest work on tool and hardware design

41 Problem Formulation Arbitrary object Parametrized transformation Contact curve: Contact vector:

42 Problem Statement Find the finger form that crosses the contact curve tangent to the contact vector

43 Part II - Solution

44 Reformulation as a Vector Field

45 Requirements and Existence Inconsistent Constraints: Strictly monotonic distance. Existence: Theorem of existence of integral curves. Feasibility: Do not penetrate the object.

46 General Solution Recipe 1. Construct contact curve and contact vector. 2. Generate vector field. 3. Integrate. Step 1 Step 2 Step 3

47 Analytic Solution Specific case: Contact curve is a line. Contact vector constant along it. Geometry normalization: Contact line vertical. Center of rotation in the X axis. Case I Case II Case III

48 3 Families of Solutions Case I Case II Case III Logarithmic spirals Spirals with limit circle Lines

49 Part III - Applications

50 Scale Invariant Grasping Gripper with contact geometry invariant to scale

51 Scale Invariant Grasping

52 Pose Invariant Grasping Gripper with contact geometry invariant to pose

53 Pose Invariant Grasping

54 Pick-up Tool Gripper to pick-up an object Preserves contact suitable for sliding

55 Rock-Climbing Cam Contact geometry invariant with wall width

56 Jar Wrench Contact geometry invariant with lid diameter

57 Anti Kickback Device Contact geometry invariant with thickness of workpiece

58 Open Problems / Future Work Finger center of rotation. Objective Combination: Pose and scale invariant? Grasp Variance. Application to Locomotion Design of feet shape. Design of open-loop trajectories.

59 Take Home Message Grasp Invariance: principle to guide the design of finger and fixture form. General Solution Recipe.

60 Thanks!!