Milli-RIO: Ego-Motion Estimation With Low-Cost Millimetre-Wave Radar

Abstract

Robust indoor ego-motion estimation has attracted significant interest in the last decades due to the fastgrowing demand for location-based services in indoor environments. Among various solutions, frequency-modulated continuous-wave (FMCW) radar sensors in millimeter-wave (MMWave) spectrum are gaining more prominence due to their intrinsic advantages such as penetration capability and high accuracy. Single-chip low-cost MMWave radar as an emerging technology provides an alternative and complementary solution for robust ego-motion estimation, making it feasible in resource-constrained platforms thanks to low-power consumption and easy system integration. In this paper, we introduce Milli-RIO, an MMWave radar-based solution making use of a single-chip low-cost radar and inertial measurement unit sensor to estimate six-degrees-of-freedom ego-motion of a moving radar. Detailed quantitative and qualitative evaluations prove that the proposed method achieves precisions on the order of few centimeters for indoor localization tasks.

Millimeter Wave Sensing: A Review of Application Pipelines and Building Blocks

“…and lat. R and v Undefined Radar [146] Doppler shift data (time, TX channel) [190] [63] Spectrogram (range, velocity) Radar [181] Radar image (x, z) Active imaging [191] Radar image (x, z) Active imaging [67] R, ϕ, and vr Undefined Radar [192] Radar image (x, z) Active imaging [193] Radar image (x, z) Active imaging [69] Heartbeat data (time) Radar [189] Radar image (x, z) Active imaging [72] Spectrogram (x, z) [194] [176] Radar image (x, y) Active imaging [177] Radar image (x, y) Active imaging Point cloud frame (x, y) R, vr , θ, and power Undefined [81] Spectrogram (range, velocity) [195] [83] Spectrogram (range, azimuth) Radar [84] Profile (range) Radar Range compressed down-conversion IF data [89] Point cloud frame (x, y, z) Radar [91] Spectrogram (range, velocity) Radar Point cloud frame (x, y, z) vr Undefined [97] Point cloud frame (x, y, z) Radar [99] R and θ Undefined Radar [196] Reflection intensity Undefined Radar Profile (range) Spectrogram (time, frequency) [104] Radar trace Undefined Radar [178] Spectrogram (x, y) Active imaging [105] Heartbeat data (time) [198] [112] Spectrogram (time, frequency) [199] [185] Positioning data Undefined Spatial sweeping [115] R and θ Undefined Radar [153] RSS signal (time, RX channel) Spatial sweeping Profile (range) Waterfall chart (time, range) Dynamic (time-varying) signal components Point cloud frame (x, y, z) vr Undefined Point cloud frame (x, y, z) vr and intensity Undefined [125] Spectrogram (range, velocity) Radar [131] Spectrogram (x, y) Radar…”

“…They require a model to translate information from pre-processed datasets or extracted feature sets to the application goal. Examples of these models include classification models [46], [47], [79], filter models [46], [89], [185], and measurement models [133], [159], [184], [186]. Classification models assign class labels to a set of input data.…”

“…Several Bayesian filters exist, such as the custom Bayesian [45], [185], particle [44], [136], α − β [128], Kalman [46], [53], [58], extended Kalman [55], [94], [96], [114], [115], [135], [186], fusion extended Kalman [95], [146], unscented Kalman [89], fusion adaptive Kalman [51], and adaptive Sage-Husa Kalman [52], [61] filter. The Kalman filter [219], under linear, quadratic, and Gaussian assumptions, can represent the state transition and observation functions x t = g(x t−1 , pn t ) and s t = h(x t , mn t ) as a set of linear equations.…”

“…In the second step, to find the optimal rotational angle and translation, the sum of squared distances between the points is used [135]. Normal Distribution Transformation (NDT) scan matching [89], [134] requires a grid of probability functions created from a map for matching.…”

“…Association/allocation models either use combinatorial optimization [46], [185], a gating function [55], [94], [96], [114], template model matching [201] or landmark association [89], [132]. These models associate input data to an existing entity or create new entities based on an optimization criterium.…”