State estimators in deployed drones
A drone, in the dictionary sense, is “an uncrewed aircraft or vessel guided by remote control or onboard computers” (Merriam-Webster). Industry and defense usage stretches the word to ground vehicles as well: uncrewed ground vehicles are routinely called ground drones (IEEE Spectrum), and uncrewed surface vessels are sold as sea drones, as in Saildrone. The table below uses the word in that broad sense and lists aerial, ground and marine systems whose published documentation names the filter they run.
The EKF dominates where an inertial sensor drives the prediction and GNSS, a barometer or a camera corrects it, which covers most autopilots. Sampled and grid filters appear where the belief can have several modes, as when a robot localizes against a map from an unknown start or an underwater vehicle matches depth soundings against a seabed map that has repeating features. HUGIN shows the two working together. The Kalman filter carries the inertial solution and drifts between fixes, and the point-mass filter, whose grid represents a multimodal belief over position, hands it a fix only once its estimate has converged.
A drone with an LTE or 5G modem could also estimate its position from cellular measurements. Ray, Deb and Monogioudis located modems in a 4G network with a particle filter that uses signal strength reports, with a median outdoor error of about 20 m (IEEE INFOCOM 2016). A 5G extension adds the indices of the beams the modem reports, and reached about 4 m in line of sight in simulation (ICNC 2019). Combined with the inertial prediction, such an estimate could stand in for GNSS when it is jammed or blocked. None of the systems in the table uses cellular measurements. The particle filter section describes the method and the changes a drone would need.
Sections
State estimation
Recursive Bayesian estimation, Bayes filters, Kalman filters, and HMM-based localization.
Recursive state estimation
The general recursive Bayes filter and its assumptions.
Discrete Bayesian filter
The Bayes filter on a finite state space, with a worked grid-world example.
Kalman filters
Optimal estimation for linear models with Gaussian noise, built up in one dimension.
Extended Kalman filter
Linearizing nonlinear motion and measurement models, and where that fails.
Markov localization
The Bayes filter with a known map, and global localization against tracking.
EKF localization
Tracking a robot’s pose against landmarks with known correspondences.
Importance sampling
Expectations under a target from samples of a proposal, with weights and resampling.
Particle filter
A sample-based Bayes filter, used to locate drone modems from LTE and 5G measurements.
Occupancy mapping
Mapping with known poses: a log-odds grid built from lidar scans, and what odometry drift does to it.
Correspondence estimation
Matching measurements to the map: data association, joint compatibility and ICP.
SLAM
Simultaneous localization and mapping: estimating pose and map together.
Found an error on this page? Report an issue.