1 / 70

Transparency Improvement of Haptic-based Networked Systems

Transparency Improvement of Haptic-based Networked Systems. Seokhee Lee. Contents. Introduction Transparency Analysis for Haptic-based Networked Systems Delay Compensation Scheme for Haptic-based NVEs Transmission and Error Control Scheme for Haptic-based NVEs

dyan
Download Presentation

Transparency Improvement of Haptic-based Networked Systems

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Transparency Improvement of Haptic-based Networked Systems Seokhee Lee

  2. Contents Introduction Transparency Analysis for Haptic-based Networked Systems Delay Compensation Scheme for Haptic-based NVEs Transmission and Error Control Scheme for Haptic-based NVEs Haptic Synchronization Scheme for Force-reflecting Teleoperation Conclusions and Future Work

  3. Introduction Overview of HNS Motivation and Contributions

  4. Overview of HNS • HNS (haptic-based networked system) • System which provides haptic feeling for a user who interacts with remote environments • Haptic-based NVE (networked virtual environment) + force-reflecting teleoperation • Local haptic system + remote haptic systems

  5. Requirements of HNS Focused requirement • Main goal ofHNS • Executing a task in a remote environments (real or virtual) • With stability and transparency • Stability • The primary requisite for safe system • Instability • Uncontrollable oscillations and chaotic behavior • Factors causing instabilities • Quantization error, time delay, packet losses, asynchronous switching between continuous- and discrete-time subsystems …… • Transparency • Transparency ≈ haptic realism • Mathematically more difficult to analyze since the ultimate goal is to make the user experience a “good feeling” • Optimal transparency • The user cannot distinguish between direct and tele-interaction with a remote environment.

  6. Network Problems Focused network problems • Instability • The delayed and lost haptic data destabilize HNS. • The instability may cause serious damages to the user. • Transparency deterioration • The human haptic sense is more sensitive than the visual and auditory senses. • Because of the sensitivity, the transparency is deteriorated severely with network variations of delay and loss. • High transmission rate • A high update rate (approximately 1 kHz) of haptic rendering leads to high transmission rate of 1000 packets/s. • The available network bandwidth for multiple haptic data may not be sufficient over current Internet.

  7. Existing Haptic Data Networking Schemes Focused networking schemes • Network layer solutions • Improve the performance of the network itself for HNS. • QoS provision for haptic data • Class-based weight fair queue (Marshall08), • DiffServ & IntServ (Hirche05) • QoS routing for haptic data • QoS-guaranteed overlay routing (Cen05) • High speed network for haptic data • Optical networks (LaMarche07) • Application and transport layer solutions • Improve the system performance on the assumption that the current network performance is not sufficient. • Delay/jitter compensation • Transmission control • Error control

  8. Motivation • Problems of the existing networking schemes (application and transport layer) • There was little consideration to enhance the human haptic perception. • Network adaptation schemes for conventional multimedia were applied to the haptic interaction without careful consideration of the difference between the haptic event and the other data. • Remaining challenge • Minimizing the transparency (haptic realism) degradation • By determining the optimum adaptation parameters for transparent haptic interactions • Transmission rate, error control level, buffering time, spring coefficient for delay compensation,……

  9. Contributions • Distinguishing point of the proposed networking schemes • Consideration of human perception characteristics (transparency) for the optimization • Transparency analysis • Deterioration of haptic interaction quality caused by network delay is quantified as distortions of mass, damping, and spring coefficients of a virtual object. • In order to formulate the importance of a haptic event with respect to a packet loss, loss effect of each haptic event is quantified for haptic-based NVEs with a prediction algorithm. • For force-reflecting teleoperations stabilized by a control algorithm, the force feedback distortions caused by network delay and packet loss are quantified when robot keeps in contact with a wall.

  10. Contributions • Networking schemes based on transparency analysis • Delay compensation scheme • Allowable delay prediction • Predicts the maximum allowable delay based on the quantified values of mass and damping distortions. • Spring coefficient modification • Modifies spring coefficient according to delay based on the quantified values of spring coefficients’ distortions. • Verification • Experimental results • The proposed scheme improves the haptic interaction quality more efficiently compared with existing delay compensation schemes while avoiding unnecessary trial-and-errors over network delay. • Representative paper • Elsevier Computer Communications, 2009

  11. Contributions • Transmission and error control scheme • Haptic event prioritization • Based on the quantified loss effect, all haptic events are classified into several groups with different network QoS requirements. • Priority-based haptic event filtering • Based on the prioritization, select more appropriate data to be transmitted for realistic haptic interaction over bandwidth-limited lossy network. • Verification • Simulation and experiment results • The proposed scheme provides lower packet rate than the existing schemes for a transparent haptic interaction over a bandwidth-limited network. • The proposed scheme guarantees less processing delay and better haptic interaction quality compared with previous error control schemes over a lossy network. • Representative paper • Springer Multimedia Systems, 2009 • Patent • Haptic event transport method for haptic-based collaborative virtual environment and system therefor

  12. Contributions • Haptic synchronization (buffering) scheme • Allowable delay and loss time prediction • Predicts maximum allowable delay and loss time based on quantified force feedback distortions and predefined transparency requirements. • Network-adaptive buffering time control • Improves transparency by controlling the playout time of the transmitted haptic event with the transparency-related parameters from transparency analysis. • Verification • Remote calligraphy system • The proposed scheme guarantees less force feedback error compared with the existing haptic synchronization schemes over time-varying network delay. • Representative paper • ACM NetGames, 2009 • Patent application • The method for synchronizing haptic data and the haptic system

  13. Transparency Analysis for Haptic-based Networked System Transparency Deterioration Caused by Network Variations Transparency Analysis for Haptic-based NVEs Transparency Analysis for EBA-based Teleoperations

  14. Transparency Deterioration Caused by Network Variations • Transparency deterioration caused by delay • If there is network delay while a user manipulates a virtual object, the virtual object does not move immediately. • During that time period, penetration depth increases. • Eventually, the force feedback in spring-damper model also increases.

  15. Transparency Deterioration Caused by Network Variations • Transparency deterioration caused by packet loss • Haptic-based NVE without a data prediction scheme • The force feedback also increases in the same manner of the delay case. • Sudden movement • Haptic-based NVE with a data prediction scheme • If the haptic event can be predicted based on past patterns of events, the quality degradation will be small. • Otherwise, the quality degradation is more severe. With data prediction scheme <Original movement of a virtual object> <Unintended movement caused by loss>

  16. Transparency Deterioration Caused by Network Variations • Transparency deterioration caused by delay jitter • More severe deterioration of haptic interaction quality • Out-of-order arrivals as well as delayed data transmission and empty sampling instances (Lee06)

  17. Transparency Analysis for HNS • Quantification of transparency deterioration according to network delay and packet loss • In order to determine the optimum adaptation parameters of haptic data networking schemes • Focused HNS • Haptic-based NVEs • With CS (client-server) architectures • Force-reflecting teleoperation • With EBA (energy bounding algorithm)

  18. Transparency Analysis for Haptic-based NVEs • Position-position interaction • HIP (haptic interaction pointer) position (client -> server) • Virtual object position (server -> client) • Consistency server • Updates movements of all virtual objects. • Distributed force calculation • Each client calculates force feedback by using spring-damper model.

  19. Transparency Analysis for Haptic-based NVEs • Transparency • Equality between the human (Zh) and the environment (Ze) impedances • Simplification • The number of clients is not related to transparency. • A virtual object is represented by a mass and damping. • Impedance

  20. Transparency Analysis for Haptic-based NVEs • Magnitude responses of impedances with respect to network delay • Confirm that if delay increases, force feedback also increases • It is called transparency (force feedback) distortion • Mass and damping distortion (mh, bh) • Spring coefficient distortion (khm, khb) kh=0.5,ke=0.5,m=0.25, b=0.0025 R=RTT (round trip time)

  21. Transparency Analysis for Haptic-based NVEs • Mass and damping distortion • If delay increases, a user perceives a virtual object as having larger mass and damping coefficients than the actual values. • Quantification of mass (mh) and damping (bh) distortion • Low frequency approximation • Human operator usually generates low frequency input (Hirche05) • [10-6~102] • Approximated human impedance • Transparency condition • Distorted mass • Distorted damping

  22. Transparency Analysis for Haptic-based NVEs • Distortion of spring coefficient • If delay exists, user is provided with unrealistically large force feedback that seems to be obtained with the larger spring coefficient than the actual value. • Quantification of spring coefficient distortion • Mass-based distortion (khm) • Damping-based distortion (khb) • If virtual object dynamics are affected by mass more than by damping, spring distortion follows mass-based distortion; otherwise it follows damping-based distortion.

  23. Transparency Analysis for Haptic-based NVEs • Force feedback distortion according to packet losses • Without a prediction scheme • Quantified in the same manner as the force distortions according to network delay • By using loss time (Tloss) • With a prediction scheme (fdis(n)) • Quantified by using the difference between the actual and the predicted (xpre(n)) positions • Loss effect (LE(n)) of the nth haptic event • In order to formulate the importance of a haptic event with respect to a packet loss

  24. Transparency Analysis for EBA-based Teleoperation • EBA (energy bounding algorithm) • Stability algorithm of a haptic simulation system (Kim04) • EBA restricts the energy generated in the ZOH (zero order hold) within a consumable energy limit in the haptic device. • Can be applied to teleoperation to ensure robust stability regardless of the amount of time delays and packet losses (Seo08).

  25. Transparency Analysis for EBA-based Teleoperation • SlaveEBA Control law Control law Bounding law Bounding law MasterEBA

  26. Transparency Analysis for EBA-based Teleoperation • Definition of Transparency • Similarity between the force feedback for a slave robot (FsEBA) and that for a user in master side (FmEBA) • Assumptions • Robot keeps in contact with a wall and a user feels the force feedback by moving haptic device facing the wall with constant velocity (vm). • Force feedback increase according to the user's input • From the control law in master EBA

  27. Transparency Analysis for EBA-based Teleoperation • Approximation of force feedback increase • Assumption: c1m≥2c2m • γm,max(n) in bounding laws converges into c2mwhen Fm(n-1) increases monotonically. • Force feedback decrease caused by delay • Network delay reduces Tinteras much as the delayed time. • If the network delay increases, the force feedback decreases in proportional to c2m∙vm.

  28. Transparency Analysis for EBA-based Teleoperation • Force feedback decrease (Fde,loss) caused by loss • Packet losses reduces Tinteras much as the loss period. • Transparent loss time (Ttr,loss): time period when the force feedback continuously increases even though there exist the packet losses

  29. Delay Compensation Schemefor Haptic-based NVEs Related Work Delay Compensation Scheme Experimental Results

  30. Related Work • Spring coefficient modification scheme (Fujimoto04) • Dynamically changes the reaction force applied to a user by adjusting a spring coefficient of a spring-damper model. • The spring coefficient (Kh) is modified according to the current end-to-end delay (ΔT). • Problems • An accurate spring coefficient for the realistic haptic interaction can only be found by a process of trial and error. • A significant challenge remains in developing an optimum method of determining the spring coefficient.

  31. Transparency Analysis-based Approach • Transparency Analysis • Transparency deterioration caused by delay is quantified as distortions of mass, damping, and spring coefficients of virtual objects. • Delay compensation scheme based on transparency analysis • Predicts the maximum allowable delay from the quantified values of mass and damping distortions. • Modifies the spring coefficient according to network delay based on the quantified value of spring coefficient distortion.

  32. Delay Compensation Scheme Delay compensation scheme

  33. Delay Compensation Scheme • Allowable delay prediction • Quantified mass and damping distortions are proportional to network delay. • Predefined transparency requirements • Maximum allowable mass (mallowable) and damping (ballowable) • Allowable mass and damping gradients (cm, cb) • Mass-based: • Damping-based:

  34. Delay Compensation Scheme • Spring coefficient modification • Estimation of the spring coefficient increment perceived by user • By using the spring coefficient distortion in transparency analysis • Modified new spring coefficient • By subtracting the spring coefficient increment from original spring coefficient

  35. Experimental Results • Verification of the transparency analysis and delay compensation scheme • Pushing motion of a virtual object • With constant velocity • Haptic devices • PHANToM Omni • Network emulation • NIST Net

  36. Experimental Results • Verification of transparency analysis • Force feedback comparison • Force feedback 1 • Original coefficients (m, b, kh) • Delay 300 ms • Force feedback 2 • Original spring coefficient • Distorted mass and damping coefficients when delay=300 ms (m=mh, b=bh) • No delay • Force feedback 3 • Original mass and damping coefficients • Distorted spring coefficient when delay=300 ms (kh=khm) • No delay • All force feedbacks are similar -> quantifications of force feedback distortions are acceptable.

  37. Experimental Results • Verification of the proposed scheme • Force feedback comparison • Force feedback 1 • No delay (transparent force feedback) • Force feedback 2 • No delay compensation scheme • Delay of 300 ms • Force feedback 3 and 4 • Existing schemes with different initial settings (Fujimoto04) • Delay of 300 ms • Force feedback 5 • Proposed scheme • Delay 300 ms Transparent force feedback • Force feedback with proposed scheme is most similar to original force feedback -> proposed delay compensation scheme is useful for transparency improvement.

  38. Transmission and Error Control Scheme for Haptic-based NVEs Related Work Haptic Prioritization Priority-based Filtering Simulation and Experiment Results

  39. Haptic-based NVE over Bandwidth-limited Lossy Network Focused transmission control • Transmission control • Transmission rate adaptation • Transmission rate reduction (traffic reduction)

  40. Related Work • Transmission rate reduction • Statistical approach • Focuses on the statistical properties of haptic signal • Applies conventional data compression schemes to haptic data • Haptic compression scheme based on Huffman coding (Hikichi01) • Haptic compression scheme based on DPCM (differential pulse code modulation) (McLaughlin02) • Problem • Even very good compression on the payload itself is useless if a big share of the necessary network bitrate is wasted by packet headers. • Perception-based approach • Packet rate reduction rather than payload compression by using the limitations of human haptic perception

  41. Related Work • Perception-based approach • Deadband-based filtering (Hirche05; Zadeh08) • Events are only sent if the change exceeds a given threshold value. • Prediction-based filtering (Kanbara04; Clarke08) • Sender transmits the event only when the difference between predicted and actual events is larger than a threshold value. • Problem • They make the haptic applications very sensitive to packet losses. • Should be used together with a error control scheme over lossy network.

  42. Related Work Teleoperation layer • Error control scheme • ARQ (automatic repeat request) • Teleoperation layer (LaMarche07) • TCP for critical haptic data • Smoothed SCTP (Dodeller2004) • Selective ARQ for last update message • FEC (forward error correction) • STRON (Cen05) • To compensate undesirable jitter caused by ARQ • Error-correcting code such as Reed-Solomon code • Problem • The existing researches use conventional error control schemes for video, audio, and events • Large processing delay for haptic interactions • Transparency deterioration e.g. robot arm control control: robot control, feedback1: force, feedback2: video, feedback3: orientation STRON (supermedia transport for teleoperations over overlay networks)

  43. Transparency Analysis-based Approach • Transparency analysis • The packet loss effect of each haptic event is formulated based on the difference between the actual and predicted positions. • Transmission and error control scheme based on transparency analysis • Haptic event prioritization • All haptic events are classified into several groups with different network QoS requirements based on the formulated values. • Priority-based haptic event filtering • Based on the prioritization, selects more appropriate data to be transmitted over bandwidth-limited lossy network. • Guarantees less processing delay than existing error control schemes. • By reconstructing lost high-priority event with received low-priority events without any error-correcting code and retransmission

  44. Haptic Prioritization • THHLI • Derived empirically • Set to 0.4 mm • A user cannot perceive the deterioration of interaction quality within the range of 0.01 to 0.4 mm (Hikichi01) • Comparison between loss effect (LE) and a threshold value (THHLI) • HLI (haptic event loss index) • HLI=0 (HLI0) • Predictable haptic event (LE(n) < THHLI) • HLI=2 (HLI2) • Unpredictable haptic event (LE(n)≥ THHLI) • Critical haptic event • HLI=1 (HLI1) • Redundant haptic event • Predictable but necessary events for the lost unpredictable events • To compensate the loss effect of HLI2event

  45. Haptic Prioritization Gilbert loss model • α • A sufficiently small value (e.g., 10-4), • Indicating that a packet loss rate of less than α is considered negligible. • Decision of HLI1 event • Because haptic events include the position every millisecond, two sequential events have similar values. • In order to compensate the loss effect of HLI2event, several haptic events following a HLI2event can be HLI1event. • Maximum number (NHLI1) of HLI1events is calculated based on Pi • Pi: the probability that a HLI2 event and i HLI1 events are all lost.

  46. Priority-based Filtering • Transmission rate reduction • HLI0 events are always filtered. • Filtering HLI1 events • When current transmission rate (r) > allowable bandwidth (R) • Filtering type • Sequential: filtering events sequentially • Alternate: filtering events alternately • Probability increments of packet losses when j HLI1 events are filtered (Pi,jsequential and Pi,jalternate) • Filtering number (j) • Probability (Pi,j) that a HLI2 event and i HLI1 events are lost when j HLI1 events are filtered • Predefined allowable packet loss probability (Pallowable)

  47. Priority-based Filtering • Haptic event reconstruction (xrec(n)) • To reduce loss of fidelity caused by lost HLI2event • Reconstruct lost nth HLI2event with HLI1events • Required buffering time (processing time) • NHLI1 (ms) • All haptic events are generated every time when haptic rendering module is updated (i.e., every millisecond). • Worst-case scenario for reconstruction of nth haptic event • Only n+NHLI1th haptic event is received successfully • Tn,n-l : The elapsed time between nth and n-lth events • Assumptions • n-lth event was received successfully • nthHLI2 event is lost but n+mth(1<m<k) HLI1 event is transmitted successfully

  48. Simulation Results • Matlab/Simulink Simulation • Verification of priority-based filtering scheme • Virtual object contact motion • A haptic device is fixed at a zero point. • A virtual object moves back and forth from -0.1 to 0. • Penetration depth 0~10 cm • Transparent force feedback • Force feedback when packet loss rate = 0 %

  49. Simulation Results • Verification of transparency with low transmission rate • Comparison between priority-based and deadband-based filterings • Loss rate 20% and allowable bandwidth 70 Kbps • Priority-based filtering • Satisfies the transmission rate and transparency requirements (error < 0.1N). • Deadband-based filtering • Threshold = 2.5 mm • Large force feedback error • Threshold = 0.7 mm • Large transmission rate

  50. Experimental Results Transparent force feedback • Verification of transparency over lossy network • Transparency comparison (peak force feedback error) • Prediction-based with no error control: 0.11N • Prediction-based with FEC: 0.3N • Prediction-based with ARQ (μ=25ms, ν=4): 0.1N • Prediction-based with ARQ (μ =4ms, ν=1): 0.104N • Priority-based: 0.014N (μ =retransmission timeout, ν=maximum retransmission)

More Related