Computing The Speckled Golfer DK Arvind and CA Bates Research Consortium in School of Informatics, University of Edinburgh (dka@inf.ed.ac.uk) BodyNets 08, Tempe AZ, 15Mar2008
Specks, Specknets, Speckled Computing Full-body, fully wireless, 3-D motion capture in real-time using network of Orient specks Biomechanics of the golf swing Video Future work and Conclusions
Specks are miniature programmable semiconductor devices which can sense, compute, and network wirelessly. Autonomous devices with rechargeable battery Energy scavenging: photovoltaic cell tuned for internal lighting Specks are assumed to be non-static and unreliable
Specknets and Tens/hundreds of specks collaborate as dense programmable network Specknet Sensory data processed collaboratively, and information extracted in situ fine-grained distributed computation Encapsulation of sensing, processing and wireless networking in a single device - new class of Information Processing Devices Enabler Technology for Ubiquitous Computing
Endow everyday objects with sensing, processing and wireless networking capabilities Link the sensory data from the physical world to the virtual world of networks of computers and the Internet Aim to bridge the physical and virtual worlds
Family of Speck devices 8-bit (med) client 32-bit (large) microserver 8-bit 5mm cube client Orient2 Speck 64 node testbed accessible over the internet
Full-body tracking of a Speckled Person visit http://www.specknet.org to view video A. Young, M. Ling and D.K. Arvind, Orient-2: A Wireless Real-time Posture Tracking System Using Local Orientation Estimation, Proc. 4th Int. Workshop on Embedded Networked Sensors (EmNets 2007), 25-26 June, Cork, Ireland, July 2007. Distributed Posture Tracking
Orient-2 Hardware Custom designed motion tracking platform 16-bit Microchip dspic processor 250kbps Chipcon CC1100 radio Freescale 3-axis accelerometer 2 Honeywell 2-axis magnetometers 3 Analog Devices MEMS rate gyroscopes 32Mbit STMicro FLASH 120mAh Li-Poly battery 36x28x11mm, 13gms All sub-systems under processor control Sensor sensitivity Power saving
Multidisciplinary Expertise Programmable Networks Radio Photonics Solar Cells Batteries Demonstrators Distributed Computing Processor Digital Signal Processing
5CubeOTS When Size Matters Sensor, Processor, Wireless Networking and Battery in 5X5X5mm
Golf Golf is a sport in which a player, using several types of clubs, hits a ball into each hole on the golf course in the lowest possible number of strokes. The first game of golf for which records survive was played at Bruntsfield Links, in Edinburgh, Scotland, in A.D. 1456 -- Wikipedia "If you watch a game, it's fun. If you play it, it's recreation. If you work at it, it's golf." -Bob Hope
Analysis of golf swing Aim. on the golf course, and not in the studio Accurate, infrastructure-less, portable, strap-on Real-time feedback either sonic or visual (on a PDA) Fully wireless, and full body (if required) Score the goodness of a swing based on existing body of research on the biomechanics of golf swing Estimate the distance of a swing from the personal best
Traditional Methods for Motion Capture are unsuitable Joint angle sensors Hinders free movement Bulky Visual Methods using cameras Suffers from occlusions Often requires time consuming post-processing Limited Coverage Expensive Motion capture suit wired
Analysis of Motor Control Skills - The Golf Swing Swing of the club Impact of the clubhead with the ball Flight of the ball towards the target
Swing of the Club Modelled as a double pendulum Arms of the golfer act as one pendulum connected to the club Club acts as another pendulum Equation of motion for a double pendulum using Newton s Laws
Model of the Swing Arms of the golfer swing about an axis that moves during the downswing Club swings about a moving axis near the wrists of the golfer Two rigid rods Arm (A) and the Club (C) Rod A: point halfway b/w shoulder to wrist
Biomechanics of the Swing Physics of Golf by T.P. Jorgensen, Springer, p9 a: hor. accl. at O Gamma: angle of rod A with the downward dir. at the start of the downswing Beta: wrist-cock angle Theta: downward angle of rod A Alpha: downward swing angle Alpha dot: angular vel. of rod A Alpha double dot: angular accl. of rod A
Swinging in the Plane Rule v start v V' n vstart vector pointing down the shaft at the start of swing vend shaft vector at the end of the swing n = vstart x vend and is normal to the swing plane v general shaft vector during the swing which does not lie in the plane v correct shaft vector which does lie in the plane defined by n α - angle between v and n, where v end cos(α) = v.n / v n
Swinging in the Plane Rule v start v V' n v end cos(α) = v.n / v n, α - angle between v and n v is obtained by taking the local down-shaft vector in the club sensor s co-ordinate system and rotating it by the current orientation of the device, to give a vector that points down the shaft in the world co-ordinate system vworld = q* x vlocal x q q quaternion specifying orientation of the device; q* is the conjugate; vworld, vector pointing down the shaft in the world co-ordinate system, and vlocal, in the local co-ordinate frame; x is the quaternion product The rule returns 1 cos(α) as a score, which is 1 for a swing perfectly within the plane, and 0 when perpendicular to the swing plane
Head Movement Rule v start v V' n Change in the orientation of the head about each axis between consecutive snapshots of the golf swing Orient speck attached to the cap the root for the body model, and any head movement is recorded relative to the motion of the body For each snapshot calculate the world direction vector that points along each axis of the headmounted device, and compare it to the previous value v end Score = (Abs(cos(δx)) + Abs(cos(δy)) + Abs(cos(δz))) /3, δx, δy, δz are changes in alignment along the x, y and z axis
Golf Statistics Top Row: Distance of swing from ideal plane Middle Row: Sine of the angle at the wrist, elbow and shoulder Bottom Row: Angle away from the plane and the 3 joint angles Distributed Posture Tracking
Tracking the Golf Swing (with real-time stats) Distributed Posture Tracking
System Overview Binary serial data containing orientation quaternions A list of frames. Each frame is a snapshot of the orientation of the devices Orient2 Basestation Orient2Comms Body Model Graphical display Each rule can update its own graph Swing in plane rule Motion scorer passes joint orientations to each rule and a score is returned Motion Scorer Head movement rule Each time the body model receives a new frame, it will call each frame update task. End users should write their own frame update task to interface with the body model Audio output Motion scorer outputs the overall score based on the weighted average of each rule Rule weightings FrameUpdateTasks
Analysis (with annotations) by the Motion Scorer Start of a good swing Start of a poor swing The overall score is the weighted average of the scores for each rule (below). The weightings are shown in the pie chart on the right. Low score in the second swing due to the club leaving the ideal plane and excessive movement of the head. In this example, only the swing in plane and head movement rules are used to calculate the overall score. Club leaves the swing plane. The head is moved quickly during the swing. Each rule can display an auxiliary graph for extra information. In this case, the angle of movement of the head about each axis is shown. The angle of each joint in the body model is updated in real time.
Liberate Motion Capture from the Studio Projects using Orient motion capture
Sensing, Computation and Collaboration at the edges to extract information locally, and effect actuation if required Specknet on the person, say a dancer Sensory data such as RPM during rotation (sensory data)) Track the movement of the limbs (sensing, collaboration) A robot mimics the actions of the dancer (sensing, collaboration and actuation) This information can be accessed and manipulated remotely over the internet Specknet the last millimeter of the Internet
Training Robots by Mimicking a Speckled Operator visit www.specknet.org to view the video Distributed Posture Tracking
SPeckled Computing Applications CEntre (SPACE) Reason for Existence is Evangelism to encourage people to experiment with the technology One step closer to the market than the Consortium. Is linked to the Consortium but at arms length, only takes releases of technology which can be supported Provides Specks Space Support Training Aim to be self funding Customers buy a package of: Lab Space Rent-a-Speck Support Time Access to archive of research. Turns potential ideas into products Customers use industrial-strength versions Support staff are dedicated No mixing of Research and Support time
Conclusion Demonstration of a fully wireless, unobtrusive, motion capture system for the 3-D analysis of golf swings with real-time feedback Future - Systematic analysis of user s experience for a set of golfers Interested in the Orient Motion Capture System? Contact dka@inf.ed.ac.uk www.specknet.org for more information
Selected References: D K Arvind,, in Proc. Nanotech2005, vol 3, pp 351-354, ISBN 0-9767985-2-2, Anaheim CA, USA, May 2005. R McNally, K J Wong, D K Arvind, A Distributed Algorithm for Logical Location Estimation in, in Proceedings of The IEEE Wireless Communications & Networking Conference 2005, USA, March 2005. K.J. Wong and D.K. Arvind, SpeckMAC: Low-power Decentralised MAC Protocols for Low Data-rate Transmissions in Specknets in Proc. Int. Workshop on Multi-hop Ad-Hoc Networks, Florence, Italy, May 2006, pp. 71-78, ACM Press. R. McNally and D.K.Arvind, A Distributed Leaderless Algorithm for Location Discovery in Specknets, Proc. Euro-Par 2007, Rennes, France, 28-31 August, Springer-Verlag. Acknowledgements This work was supported in part by the Scottish Funding Council under the Strategic Research Development Grant #R37329, and th Engineering and Physical Sciences Research Council under the Basic Technology Programme Grant #C523881.