Using humans to analyze robot hand capabilities John Morrow HF Seminar 5/18/2018
Quick History of Robot Hands Morrow - Evaluating Hands 2 [Graphic borrowed from RI Seminar given by Matei Ciocarlie - https://www.youtube.com/watch?v=wiTQ6qOR8o4]
Quick History of Robot Hands Morrow - Evaluating Hands 3
Fully-Actuated Hands Morrow - Evaluating Hands 4 [Jacobsen et al, 1986; Bae et al, 2011]
Quick History of Robot Hands Morrow - Evaluating Hands 5
Quick History of Robot Hands Morrow - Evaluating Hands 6
Under-Actuated Hands Morrow - Evaluating Hands 7 https://www.youtube.com/watch?v=BMLJBVPb7qM
Under-Actuated Hands Morrow - Evaluating Hands 8 [Odhner et al, 2012]
Where do we go from here? Morrow - Evaluating Hands 9
Our Questions: What is the most effective addition we can make to our robot hands? How can we evaluate these hands? Morrow - Evaluating Hands 10
What additions are others making? Morrow - Evaluating Hands 11 [Ma and Dollar, 2016; Odhner et al, 2012; Aukes et al, 2014]
Versatility Morrow - Evaluating Hands 12
Physical Human Interactive Guidance Morrow - Evaluating Hands 13 [Balasubramanian et al, 2012]
The Power of Human Grasping 77% vs. 97% Robots Humans Morrow - Evaluating Hands 14 [Balasubramanian et al, 2012]
The Power of Human Grasping 93% vs. 97% Robots Humans Morrow - Evaluating Hands 15 [Balasubramanian et al, 2012]
Physical Human Interactive Guidance Morrow - Evaluating Hands 16 [Balasubramanian et al, 2012]
Our Study ● 18 subjects, 2 hands per subject ● 10 minute warm-up ● Two tasks: – Drawing with a pen – Spraying a spray bottle ● Comparing hands that were... – Under-Actuated – Fully-Actuated – Fully-Actuated and Compliant Morrow - Evaluating Hands 17
Barrett Hand (BH) ● Under-actuated ● Controlled via sliders ● Limited joint movement – Coupled per finger Morrow - Evaluating Hands 18 [Townsend et al, 2000]
Posable Barrett Hand (PH) ● Fully-actuated ● Controlled by hand ● No limitations on joint pose Morrow - Evaluating Hands 19
OpenHand Model O (OH) ● Fully-actuated ● Controlled by hand ● Compliant joints can twist Morrow - Evaluating Hands 20
Pen Task Morrow - Evaluating Hands 21
Spray Task Morrow - Evaluating Hands 23
Results Metric Task BH PH OH Avg. Task Bowl 191 256 85 Completion (sec) Spray 398 344 163 Avg Manipulation Bowl 30% 21% 22% Time (%) Spray 23% 22% 21% Avg Attempted Bowl 3 6 2 Grasps Spray 4 5 2 Morrow - Evaluating Hands 24
Results Morrow - Evaluating Hands 25
Results Morrow - Evaluating Hands 26
Hand Comparisons Metric Task BH PH OH Avg. Task Bowl 191 256 85 Completion (sec) Spray 398 344 163 Avg Manipulation Bowl 30% 21% 22% Time (%) Spray 23% 22% 21% Avg Attempted Bowl 3 6 2 Grasps Spray 4 5 2 Morrow - Evaluating Hands 27
Survey Data Morrow - Evaluating Hands 28
What is the PH missing? Morrow - Evaluating Hands 29
Study Limitations ● No data for OH ● Finger ‘loops’ difficult to use ● ‘Powered’ by humans – Not the same interface Morrow - Evaluating Hands 30
Our Questions: What is the most effective addition we can make to our robot hands? How can we evaluate these hands? Morrow - Evaluating Hands 31
Conclusions ● Refined control of distal link – Bending it backwards – Twisting it ● Evaluating hands with humans as the guide ● Observations: – Soft finger pads are important to us – Morrow - Evaluating Hands 32
References Bae, Ji-Hun, et al. "Development of a low cost anthropomorphic robot hand with high capability." IEEE/RSJ International ● Conference on Intelligent Robots and Systems (IROS), 2012 . Jacobsen, Steve, et al. "Design of the Utah/MIT dextrous hand." IEEE International Conference on Robotics and Automation. ● Proceedings. Vol. 3. IEEE, 1986. Lael U. Odhner, Raymond R. Ma, and Aaron M. Dollar, "Precision grasping and manipulation of small objects from flat surfaces ● using underactuated fingers," 2012 IEEE International Conference on Robotics and Automation (ICRA), pp.2830-2835, 14-18 May 2012. Raymond R. Ma and Aaron M. Dollar, "In-Hand Manipulation Primitives for a Minimal, Underactuated Gripper with Active ● Surfaces," ASME International Design Engineering Technical Conferences (IDETC), 2016. Aukes, Daniel M., et al. "Design and testing of a selectively compliant underactuated hand." The International Journal of ● Robotics Research 33.5 (2014): 721-735. Balasubramanian, Ravi, et al. "Physical human interactive guidance: Identifying grasping principles from human-planned ● grasps." IEEE Transactions on Robotics 28.4 (2012): 899-910. Townsend, W. T. "MCB—industrial robot feature article—Barrett hand grasper." Industrial Robot: An International Journal 27.3 ● (2000): 181-188. Morrow - Evaluating Hands 33
Expertise modeling and training: Manual 3D Image Segmentation Process Cindy Grimm Ruth West (UNT) Chris Sanchez (OSU) Anahita Sanandaji (PhD) Max Parola, Meghan Kajihara, Deniece Yates, Brandon Lane
Problem area and goals • There are many tasks that require a mix of spatial ability , domain knowledge, verbal skills, and mathematical skills o Mechanics of solid materials o 2D/3D image segmentation o Biological processes • How do we transfer (spatial) knowledge from experts to novices? o Capture and representation of expert’s mental models o Decomposition into hierarchical and orthogonal skills o Training materials • Tools o [Capture] Eye-tracking, EMG, video, spatial relationships o [Training] AI, Virtual/Augmented reality, computer interfaces 1
3D Image Segmentation § A fundamental process in: • Scientific and medical applications § Medical Imaging and Segmentation • Locating tumors • Measuring tissue volumes • Computer guided surgery § Performed (or evaluated) on 2D slices of the 3D data • Stack of CT Scans 2 Stack of CT scan of a liver
3D Image Segmentation Approach § Drawing contours on selected cross-sections by human experts 7
3D Image Segmentation Approach § Drawing contours on selected cross-sections by human experts Contour s Selected cross-section of a developing chicken heart 8
Time-intensive Process § Performing segmentation manually on a slice-by-slice basis Manual segmentation by Reconstruction human experts 9
Expertise: What does it consist of? § Domain knowledge – expected shapes/patterns/shape relationships § Spatial skills • Relationship of 2D slices to 3D shape • Relationship of image properties to contours • Software interface: Amira, etc, are big, complicated pieces of software 10
Expertise: How do we capture it? § Field studies: In-depth and per-expert analysis • Eye-tracking (spatial data) • Video and audio recoding • Task analysis • Retrospective think aloud 11
Analysis Methodology: Combining spatial/visual data with task structure • Primary source of analysis : Observation data (gaze and actions) • Match to: Participant’s mental task model 12
Analysis: Conceptual task -> gaze and actions Same site Action Pairs: Novice (P3) vs Expert (P5) 1 novice Origin to Subsequent action pairs for 1 expert expert vs novice for the task Annotate Same task Cell Navigation Navigation w a r D E w d a i t r D E d i t P5 P3 Expert Novice 13
Repeated Task Results 20
Task structures ST-L1 P7 TASKS ST-L2 ST-L1 P6 TASKS ST-L2 ST-L1 P5 TASKS ST-L3 ST-L2 ST-L1 P4 TASKS 21
Insights and Hypothesis § Comparing novices to experts: • Have better knowledge of 3D structures • Tend to use 3D views more frequently § Mental model hypothesis: • Given a 3D structure and slicing plane experts can: 1. Predict the 2D contour 2. Predict how 2D contour changes 3. Identify invalid 2D contours 4. Image characteristics that correspond to boundaries 22
Recommend
More recommend
