Project Soli

Project Soli
Basic Info
Type	Input Device, Gesture Tracker
Subtype	Hands/Fingers Tracking
Developer	Google (Advanced Technology and Projects)
Manufacturer	Infineon Technologies (FMCW research chip); SiBeam (CMOS research chip)
Announcement Date	29 May 2015 (Google I/O)
Release Date	24 October 2019 (first consumer device, Pixel 4)
Display
Precision	Sub-millimeter (0.43 mm average RMS displacement error in a 2016 laboratory test)
Tracking
Update Rate	Over 10,000 frames per second (2016 research paper); up to 18,000 frames per second (2016 redesigned chip)
Device
Sensors	60 GHz radar; 2 transmit and 4 receive antennas (2016 research chips)
Size	12 x 12 mm (FMCW chip) and 9 x 9 mm (CMOS chip), 2016 research chips

Project Soli is an input technology developed by Google's Advanced Technology and Projects (ATAP) group. It utilizes radar to detect the movements and gestures of a person's hands and fingers. Shaped as a chip smaller than a quarter, Soli can be embedded in most wearables and electronic devices. Google initially targeted wearable devices such as smartwatches.

Project Soli was revealed during Google I/O 2015 on May 29, 2015. The sensor operates in the 60 GHz millimeter-wave band, and the team's 2016 research paper describes it as tracking gestures with sub-millimeter accuracy at more than 10,000 frames per second on embedded hardware. The paper lists virtual reality and mobile VR and augmented reality systems among the intended applications.

Soli reached consumers in the Google Pixel 4 smartphone in October 2019, where it powered a feature called Motion Sense. Google left it out of the Pixel 5 but continued to use the sensor in Nest smart home devices, including the second-generation Nest Hub and the Nest Thermostat.

Reviewed 27 September 2026. Checked every new or changed claim against the 2016 Soli paper (ACM TOG), the FCC waiver order DA 18-1308, Google blog and research posts, and the TechCrunch, 9to5Google, VentureBeat, Android Authority, Android Police, New Atlas and Microwave Journal articles cited. About review dates.

History

Announcement and developer kits (2015)

ATAP presented Soli at Google I/O 2015 in San Francisco, alongside the smart textile effort Project Jacquard. Ivan Poupyrev, a technical project lead at ATAP, demonstrated it by turning a virtual knob with small finger movements above the radar. TechCrunch reported that the team had shrunk its radar hardware from roughly the size of a games console to something smaller than a quarter over about 10 months, that an API would give developers all of the translated signal information, and that Soli was targeted for broad availability later that year in a form usable in wearables such as smartwatches.

In 2015 ATAP shipped a first Soli developer kit to about 60 developers. One group that used the developer kit was a team at the University of St Andrews, whose RadarCat system used the Soli chip and machine learning to identify objects and materials placed on the sensor.

Infineon chip and prototypes (2016)

At Google I/O 2016, ATAP showed a Soli chip redesigned with Infineon. According to TechCrunch, the new chip was a third of the size of the earlier version, cut power consumption from 1.2 W to 0.054 W, and could still run at 18,000 frames per second. ATAP demonstrated it in two prototypes: a smartwatch built with LG, on which broad hand movements scrolled through messages and finer motions worked as virtual controls, and a JBL speaker that could be controlled with gestures from up to 15 meters away. ATAP said it planned a beta developer kit for the following year. Infineon's announcement of the partnership described the 60 GHz radar as "a key technology enabling Augmented Reality".

The same year, the Soli team published a detailed description of the system in ACM Transactions on Graphics and presented it at SIGGRAPH 2016.

FCC waiver (2018)

On 7 March 2018 Google asked the U.S. Federal Communications Commission (FCC) for a waiver allowing Soli sensors to operate in the 57-64 GHz band at higher power levels than the commission's rules permitted. Google argued that the existing limits were too restrictive for Soli to work well and led to missed motions. The FCC granted the waiver on 31 December 2018, using parameters from a joint filing by Google and Facebook: +10 dBm peak transmitter output power, +13 dBm peak EIRP, +13 dBm/MHz peak power spectral density, and a transmit duty cycle of no more than 10 percent in any 33 ms interval. The order also allowed Soli devices to be used aboard aircraft. The FCC said the technology could benefit users with mobility, speech or tactile impairments.

Pixel 4 and Motion Sense (2019)

Google confirmed in July 2019 that Soli would be built into the Pixel 4, writing that ATAP had been working on the radar "for the past five years". The company said Soli sensor data would be processed on the phone and not saved or shared with other Google services. The Pixel 4 and Pixel 4 XL were announced on 15 October 2019 and shipped on 24 October 2019. Motion Sense detected a hand reaching for the phone to start face unlock, turned the screen off when nobody was around, and let users skip songs, snooze alarms, dismiss timers or silence the ringer with a wave. At launch the feature was enabled everywhere the phone was sold except Japan.

In a 2020 technical overview, ATAP engineers Jaime Lien and Nicholas Gillian described the Pixel 4 module as "the first integrated short-range radar sensor in a consumer smartphone". It used a 60 GHz frequency-modulated signal, three receivers and a 5.0 mm x 6.5 mm x 0.873 mm chip package. The machine learning models were built with TensorFlow, trained on millions of gestures recorded from thousands of Google volunteers, and ran on the phone's low-power digital signal processor. Google also described the phone's antenna arrangement as one transmitter and three receivers.

Removal from phones and move to Nest devices (2020-2024)

The Pixel 5, released in 2020, did not include Soli. Android Authority reported that the Pixel 4 could not be sold in many countries, including India, because the 60 GHz band Soli needs is not open for commercial use there, and that Google hardware chief Rick Osterloh said Soli and Motion Sense would return in future products without saying whether they would be phones.

The Nest Thermostat announced in October 2020 contained a Soli module with a 60 GHz radar. VentureBeat reported that the module was developed jointly by ATAP and the Nest team, that its models ran on the thermostat's low-power processor, and that the thermostat started at US$129 in the United States. The sensor was used to wake the thermostat's screen.

In March 2021 Google announced the second-generation Nest Hub, priced at US$99.99, with a Soli-based Sleep Sensing feature and Quick Gestures such as an air tap to play or pause media and a wave to snooze alarms. Google described the Nest Hub sensor as a millimeter-wave frequency-modulated continuous wave (FMCW) radar transceiver that emits an ultra-low power radio wave. Its sleep model was trained on more than a million hours of radar data from thousands of people.

The fourth-generation Nest Learning Thermostat, launched in the United States and Canada on 20 August 2024, uses Soli for a feature called Dynamic Farsight, which shows information when it senses someone nearby. Android Police reported at the time that none of the rumors about the upcoming Pixel 9 series mentioned Soli.

Research on nonverbal cues (2022)

In November 2022 ATAP designers Lauren Bedal and Leonardo Giusti described research using Soli to read nonverbal cues such as a person's proximity to a device and whether they are turning toward or away from it. They wrote that the overlap between spatial "fields" around people and devices "is a good indicator of the level of interest between a user and a device", and noted that because the radar is not a camera and cannot identify people, it could offer a privacy-preserving approach in the home.

How it works

Soli is a 60 GHz radar that illuminates the whole hand with a single broad antenna beam instead of trying to resolve the hand spatially with a narrow beam. From the reflected signal it computes a set of hardware-independent representations, including I/Q, range-Doppler, range profile, micro-Doppler, fast time spectrogram and a three-dimensional spatial profile. Using this information, the system is able to determine properties of the hand such as its motion and velocity. Soli uses machine learning to translate these movements into pre-programmed commands; the 2016 paper used a Random Forest classifier followed by a Bayesian filter.

Soli functions in the 60 GHz radar band. The 2016 paper reports tracking at more than 10,000 frames per second on embedded hardware, and TechCrunch reported 18,000 frames per second for the chip redesigned with Infineon.

At the 2015 announcement Google said an API would give developers the translated signal information, and a first developer kit went to about 60 developers that year.

Research chips

The 2016 paper describes early prototypes, including a 57-64 GHz FMCW radar built from discrete components around Infineon's BGT60 communication chip, followed by two fully integrated Soli chips.

Chip	Size	Process and modulation	Manufacturer
FMCW chip	12 x 12 mm	SiGe, frequency-modulated continuous wave (FMCW)	Infineon
DSSS chip	9 x 9 mm	CMOS, direct-sequence spread spectrum (DSSS), on-die antennas	SiBeam

Both chips operate in the 60 GHz band, which the authors say allows small antennas and use of 7 GHz of bandwidth. Each has an antenna-in-package patch array forming a broad 150-degree beam, and two transmit and four receive antennas for digital beamforming and three-dimensional tracking. The paper gives the chips' power consumption as 300 mW with adaptive duty cycling.

In a test against a laser range finder, with a metal plate moved by a robotic arm at twelve velocities from 20 mm/s to 200 mm/s, the average RMS displacement error was 0.43 mm. The Soli processing pipeline was benchmarked on embedded platforms including a Raspberry Pi 2 and a Qualcomm Snapdragon 400.

Interaction design

Soli's gesture language is built around "Virtual Tools": a Virtual Button (pressing the index finger and thumb together), a Virtual Slider (moving the thumb along the side of the index finger) and a Virtual Dial (turning an invisible dial between thumb and index finger). Because the hand touches itself during these gestures, the authors argue that the interactions feel tactile even though no physical control or active haptics are present, which they contrast with camera-based touchless systems.

In an evaluation with five participants performing four gestures (Virtual Button, Virtual Slider, horizontal swipe and vertical swipe), the pipeline reached a per-gesture accuracy of 92.10 percent over 1,000 test gestures after Bayesian filtering, and 86.90 percent from the raw classifier output.

A later ATAP paper presented at CHI 2021, RadarNet, described a convolutional neural network that recognizes four directional swipes and an omni-swipe with a 6.5 x 5.0 mm radar chip integrated into a mobile phone. The authors report that its model size and inference time are less than 1/5000 of an earlier radar gesture recognition technique, and evaluated it on 558,000 gesture samples and 3,920,000 negative samples.

Features

• Translates hand gestures and finger movements into commands for smart devices.

• The sensor can track sub-millimeter motions at high speed and accuracy.

• Small gestures such as turning a virtual knob or dial, or sliding the thumb along a finger, can be used as controls.

• Works through materials, so the sensor can be embedded inside devices.

Products using Soli

Product	Year	Soli use
Google Pixel 4 and Pixel 4 XL	2019	Motion Sense: reach detection for face unlock, presence detection, swipe and wave gestures
Nest Thermostat	2020	Screen wake when someone approaches
Nest Hub (2nd generation)	2021	Sleep Sensing and Quick Gestures
Nest Learning Thermostat (4th generation)	2024	Dynamic Farsight presence sensing

The LG smartwatch and JBL speaker shown at Google I/O 2016 were prototypes.

Relevance to VR and AR

The 2016 Soli paper names virtual reality among the uses of the sensor in its introduction, and lists mobile VR and AR systems, game controls and wearables among possible applications. It compares radar with computer vision approaches to hand tracking, which it says still suffer from latency and difficulty with overlapping fingers, and with depth cameras, which it says improve precision but do not scale well to the small form factor of wearable and mobile devices. Infineon's 2016 announcement said that virtual reality users had lacked a way to interact with virtual content and presented the Soli radar as a technology for augmented reality.

Soli differs from camera-based hand tracking. It does not build a skeletal model of the hand; the paper says the radar features it extracts are insufficient to reconstruct the hand's skeletal structure, and the system instead recognizes motion patterns and gestures from radar signal features.

Reception

Regulation of the 60 GHz band limited where Soli could be used; Android Authority reported that this kept the Pixel 4 out of many markets, including India. Writing in 2024, Android Police said Motion Sense on the Pixel 4 had persistent functionality problems that software updates did not fully fix.

See also

• Gesture recognition
• Hand tracking
• Leap Motion
• Myo
• Meta Neural Band