Source-linked AI summary
Position Tracking for Virtual Reality Using Commodity WiFi
Manikanta Kotaru, Sachin Katti
TL;DR
VR tracking can require dedicated infrastructure or offer limited position tracking. WiCapture uses commodity WiFi and CSI to track position, reporting a median trajectory error of 0.88 cm compared with SpotFi’s 132 cm.
Problem
VR tracking can require dedicated external infrastructure, while inside-out vision systems add headset complexity and cannot track lightweight hand controllers.
Source: §1 · p. 1Method
WiCapture mines CSI from standard WiFi packets received by access points, using phase changes between packets to estimate headset displacement and compensate for signal distortions.
Source: §1 · pp. 1–2Results
0.88 cm median trajectory error for WiCapture compared with SpotFi’s 132 cm median trajectory error in the indoor office deployment.
Source: §5.3.1 · p. 7Takeaways & Limitations
The authors conclude that commodity WiFi-based position tracking could potentially enable VR on mobile devices and applications built on motion tracking.
Source: §6 · p. 8The prototype has higher latency because tracking is computed in the network, and it is less accurate than current outside-in position tracking systems.
Source: §1.2 · p. 2
Abstract
from arXiv · showhide
Today, experiencing virtual reality (VR) is a cumbersome experience which either requires dedicated infrastructure like infrared cameras to track the headset and hand-motion controllers (e.g., Oculus Rift, HTC Vive), or provides only 3-DoF (Degrees of Freedom) tracking which severely limits the user experience (e.g., Samsung Gear). To truly enable VR everywhere, we need position tracking to be available as a ubiquitous service. This paper presents WiCapture, a novel approach which leverages commodity WiFi infrastructure, which is ubiquitous today, for tracking purposes. We prototype WiCapture using off-the-shelf WiFi radios and show that it achieves an accuracy of 0.88 cm compared to sophisticated infrared based tracking systems like the Oculus, while providing much higher range, resistance to occlusion, ubiquity and ease of deployment.
1. Introduction
WiCapture uses commodity WiFi packets and CSI to track VR headset position while addressing infrastructure, occlusion, and range limits of existing systems. Its prototype reports 0.88 cm tracking error, with higher latency and lower accuracy than outside-in systems.
1. Introduction
- Existing outside-in systems need specialized infrastructure, can fail under occlusion, and typically cover about 2 m; inside-out vision systems cannot track lightweight hand controllers.Visual tracking can also fail around transparent or texture-less objects.Source: §1 · p. 1
- WiCapture derives headset position from CSI metadata in standard WiFi packets received by commodity access points.It tracks position through changes in CSI phase between packets, compensating for multipath and clock-related distortions.Source: §1 · pp. 1–2
1.2. Limitations
- WiCapture uses commodity APs, works through occlusion and across rooms, and needs only a standard WiFi chip on the headset.Its current prototype computes tracking in the network, increasing latency, and remains less accurate than outside-in systems.Source: §1 · pp. 1–2, §1.2 · p. 2
1.1. Contributions
- 0.88 cm position tracking error was demonstrated with commodity Intel 5300 WiFi chips, alongside 0.25 mm precision.The prototype has higher latency and is less accurate than current outside-in tracking systems.Source: §1.1 · p. 2, §1.2 · p. 2
2. Related work
Related systems trade range, infrastructure needs, and robustness across optical, magnetic, RF, and visual-inertial approaches. WiCapture instead estimates relative trajectories from WiFi CSI, using multiple packets to improve multipath estimation.
2. Related work
- Infrared and laser tracking can require line of sight or synchronization, while magnetic, RFID, and UWB systems trade occlusion resistance or accuracy for shorter range or specialized infrastructure.WiFi signals can pass through typical obstructions such as walls and people.Source: §2 · p. 2
- Visual-inertial systems can track camera motion but fail with transparent or texture-less scenes or when the camera is occluded; WiCapture and infrastructure-free systems may complement each other.IMUs alone track orientation, not position.Source: §2 · p. 2
- Unlike prior multipath approaches using multiple frequencies, WiCapture uses CSI from multiple packets, relying on path directions remaining stationary over short intervals.It compensates for both time and frequency offsets between source and receiver.Source: §2 · pp. 2–3
- WiCapture targets relative trajectories, whereas WiFi RSSI, ToF, and AoA systems estimate location with reported errors from meters to decimeters.WiFi tracking systems described here classify trajectories into four known gestures.Source: §2 · p. 3
3. Preliminaries
WiFi receivers derive CSI from transmitted reference signals, and the CSI reflects a multipath channel whose phase changes with transmitter motion. Packet-to-packet clock offsets also change phase, by much more than the motion signal in the example given.
3. Preliminaries
- WiFi packets begin with a reference signal that the receiver correlates and samples to calculate Channel State Information (CSI).The paper compares this CSI formation process to image formation in a time-of-flight camera.Source: §3 · p. 3
3.1. CSI calculation
- A path’s angle of departure determines the phase differences across antennas, while its attenuation contributes to the received channel.The three-antenna array illustrates the additional distance and phase at the second antenna; the model can extend to arbitrary array geometries.Source: §3.2 · p. 3, Figure 3 · p. 3, §3.1 · p. 3
3.2. Wireless channel modeling
- The received wireless channel is the superposition of paths, each represented by a steering vector and a complex attenuation.The steering matrix collects one vector per path, and the attenuation vector collects the corresponding path weights.Source: §3.2 · p. 3
3.3. Change in channel due to displacement
- Transmitter displacement changes each path’s length and phase, thereby changing the channel between consecutive packets.The path-length change depends on displacement projected along that path’s direction of departure.Source: §3.3 · p. 4
3.4. Phase distortion due to frequency offset
- A 10 ms packet gap can produce 630 radians of phase change from clock offset, versus less than 0.6 radians from 5 mm of motion.The receiver and transmitter clocks drift, so their relative phase changes between packets.Source: §3.4 · p. 4
4. Design
WiCapture estimates path directions from multiple CSI packets, then uses changes in path attenuation to estimate transmitter displacement. Comparing phase changes across paths removes their shared frequency-offset term, and estimates from multiple access points can be combined.
4. Design
- Each access point sends CSI to a central server, which estimates path directions and attenuations before inferring transmitter displacement from consecutive packets.The design describes these as the two main processing steps.Source: §4 · p. 4
4.1. Estimating AoD of all the paths
- CSI from packets with nearly unchanged path directions provides diverse linear combinations for MUSIC to estimate the shared steering vectors and path directions.Small transmitter motion can substantially change path weights without measurably changing the steering matrix.Source: §4.1 · pp. 4–5
4.2. Estimating the displacement of the transmitter between consecutive packets
- Comparing phase changes across paths cancels their shared frequency-offset term, allowing displacement to be estimated from path directions and attenuation changes.The least-squares system can combine matrices and vectors from multiple access points because the target displacement is shared.Source: §4.2 · pp. 5–6
5. Evaluation
WiCapture’s stationary-position jitter was compared with Oculus DK2, while the evaluation used off-the-shelf WiFi hardware and a four-AP setup.
5. Evaluation
- The evaluation used Intel 5300 WiFi cards, four APs, and a 167 Hz trajectory update rate.CSI was calculated every 6 ms; the target had a three-antenna circular array.Source: §5 · p. 6
5.1. Stationary experiments
- Stationary jitter is the standard deviation of the estimated trajectory, relevant because the VR scene should not change when the user is still.The target transmitted 1,000 packets while stationary.Source: §5.1 · p. 6
- 0.25 mm median jitter for WiCapture compares with 0.10 mm for Oculus DK2, measured at the same target positions.WiCapture’s median is over 21 experiments at different target positions; Oculus’s camera was 0.75 m from the headset.Source: §5.1 · p. 6
5.2. Controlled tracking experiments
Controlled mechanical movements tested WiCapture’s position-estimation error and its ability to resolve small target motion.
5.2. Controlled tracking experiments
- 0.11 cm was the maximum position-estimation error as the target moved in 0.1 cm increments and decrements.The target transmitted one WiFi packet at each position.Source: §5.2 · p. 6
- 0.27 cm was the maximum error on a mechanically positioned trajectory, while WiCapture resolved millimeter-level target motion.The trajectory is shown in Fig. 6(b).Source: §5.2 · p. 6
5.3. Extensive tracking experiments
Extensive tracking tests covered indoor, occluded, and outdoor deployments, comparing WiCapture’s trajectory errors with SpotFi where reported.
5.3. Extensive tracking experiments
- The study compared WiCapture with SpotFi but did not compare with ultrasound because of its dedicated-infrastructure requirements.The paper describes the tracking accuracy as dependent on multipath, wall material, metallic objects, and AP density.Source: §5.3 · pp. 6–7
5.3.1 Indoor office deployment
- 0.88 cm median trajectory error for WiCapture compared with 132 cm for SpotFi in the indoor office deployment.The study traced 97 trajectories in a 5 m × 6 m room with APs at the four corners.Source: Figure 8 · p. 7, §5.3.1 · p. 7
5.3.3 Outdoor deployment
- 0.85 cm median trajectory error for WiCapture compared with 57 cm for SpotFi outdoors.The outdoor deployment covered a 5 m × 6 m space and traced 89 trajectories.Source: Figure 8 · p. 7, §5.3.3 · p. 8
5.3.2 Occlusion deployment
- 1.51 cm median trajectory error for WiCapture when all APs were occluded by furniture or walls.The study traced 64 trajectories with one or two APs outside the target’s room.Source: Figure 8 · p. 7, §5.3.2 · pp. 7–8
5.3. Extensive tracking experiments
- The evaluation used Oculus DK2 as ground truth and compared trajectories after translation and rotation, without scaling.The comparison metric was absolute point-by-point position difference after alignment.Source: §5.3 · p. 7
5.3.2 Occlusion deployment
- Three or more APs outside the room led to large trajectory errors; accurate tracking required at least two APs not separated from the target by walls.The authors identify this as a deployment boundary for WiCapture.Source: §5.3.2 · p. 8
5.4. Deep dive into WiCapture
The deep dive tested two WiCapture techniques: estimating AoD from multiple packets and correcting phase distortion from frequency offset.
5.4. Deep dive into WiCapture
- The experiments separately tested multipath resolution using multiple packets and phase-distortion removal for frequency offset.Both tests used data from the indoor office deployment.Source: §5.4 · p. 8
5.4.1 Improved AoD estimation
- 11 degrees was the 80th-percentile AoD estimation error using multiple packets, 3× smaller than with a single packet.The comparison used the estimated AoD closest to the manually measured ground-truth direct-path AoD.Source: §5.4.1 · p. 8
5.4.2 Removing phase distortion from frequency offset Method:
- Ignoring frequency offset made trajectory estimation errors 6× worse than WiCapture’s approach.The alternate “assume same clock” method ignored frequency offset.Source: §5.4.2 · p. 8
6. Discussion and conclusion
The discussion presents WiCapture as a commodity-WiFi tracking system with potential to enable VR on mobile devices, while noting its positioning-only focus and possible extensions.
6. Discussion and conclusion
- Atheros, Intel, and Marvell WiFi chip families expose CSI, leading the authors to believe WiCapture can be added to commodity WiFi infrastructure.Source: §6 · p. 8
- Additional WiFi frequencies and receiver antennas could improve AoD estimation, while 3D tracking could extend the algorithm's path-direction estimation into three dimensions.Source: §6 · p. 8
- The paper addresses positioning alone; orientation can be obtained relatively easily from IMUs already available on many devices.Source: §6 · p. 8
- WiCapture was developed and implemented, with performance demonstrated across different deployment scenarios.The authors say a commodity WiFi-based position tracking system could potentially enable VR on mobile devices and applications built on motion tracking.Source: §6 · p. 8
Glossary
The glossary defines WiFi tracking terms and basic communication concepts used in the paper.
Glossary
- AoD is the angle of a path relative to the line joining two antennas at the source.Source: Glossary · p. 8
- AP means a WiFi access point or router.Source: Glossary · p. 8
- CSI is the received signal when the source transmits the reference signal.Source: Glossary · p. 8
- Frequency offset is the difference between the reference-signal frequencies used at two WiFi chips.Source: Glossary · p. 8
- A packet is a WiFi unit of data sized for efficient communication.Source: Glossary · p. 8