Research

Research Topics

The lab excels in various research competencies, exercised in numerous past and ongoing projects in two general research directions.

Position and attitude determination

Position and attitude determination of moving platforms or subjects is the main-stream of the lab research activity. The laboratory expertise in algorithm development for real-time or post-mission positioning have been applied to vehicle and pedestrian navigation and trajectography. We make use of satellite based (GPS, Glonass) positioning, inertial sensors, magnetic sensors, imagery and lately the networked based positioning via Ultra-wide band or 802.x technology.

Mapping and remote sensing 

Sensor integration and close-range remote sensing competences served in development of task specific or general mobile mapping systems. Geodesy, surveying and cartography are the traditional proficiencies present in the laboratory. High precision surveying by satellite or terrestrial observations as well as network design and adjustment remains the valuable expertise that is regularly made available for consultation or research-mandates.

Projects

Current Projects

The objective of the GAMMS H2020 project is to develop autonomous mapping systems capable of producing high-definition maps. A major challenge in urban mapping scenarios is the estimation of the position of the vehicle during conditions when GNSS signals are not available. The role of EPFL in this project is to tackle this challenge by bringing in an innovative method that mathematically constrains the estimation of vehicle motion. These constraints are encapsulated using a synthetic sensor based on the Vehicle Dynamic Model (VDM). 

This project aims to enhance the accuracy and reliability of point-clouds from mobile laser scanners by precisely integrating LiDAR, visual, and inertial measurements, even in the presence of errors, and streamlining the processing pipeline.

Past Projects

While the expansion of drone applications seems to grow without boundaries, their implementation is conditioned by operational safety. Related to that is the drone ability to maintain its intended trajectory irrespective to the disturbances in received signals through navigation autonomy. In times when drones navigation relies primary on satellite positioning (e.g. GPS), the fragility of satellite signal reception poses serious concerns in safety. Hence, there is an upswing in proposing backup plans that are mostly based on optical sensing. Nevertheless, limited visibility, resolution, or surface texture, can provide challenges for such visual based navigation systems.
 
Two parallel and complementary approaches of implementing VDM techniques are followed. The first one attempts to minimise the need of additional to classical sensor(s) (IMU, velocity measurements and control estimates). This will aim to make the approach more general rather than specific thus adopt it for low cost and existing drone configurations. The second approach is aimed for future high-end drones and EVTOL aircrafts. Having the designer freedom to incorporate in the drone design Flush Air Data System (FADS). An array of several sensors measuring states at the boundary layer of a drone. The sensor array signals are then regressed by novel physics guided machine learning algorithms in order to estimate the aerodynamic coefficients or forces directly.
 
These projects enable creation of a new technological platforms for navigation of drones that address the aspects of self-localization safety. The methodologies enable safe operation in GPS-denied environment by a novel way of using interoceptive means. This shall future proof and extend current market share in accordance with envisioned legislative evolution around the world.
 
Sponsors
 
The general development of the project is sponsored by Swiss National Science Foundation, Armasuisse, Swiss Commission for Innovation (InnoSuisse) and Swiss Data Science Centre



We combine modern aircraft-based sensors with deep-learning data analysis and in-situ observations to generate spatially explicit inventories of species in the treeline  ecotone within one Alpine valley. When validated, the approach is intended to be extended over broad regions where in-situ measurements are not feasible and satellite imagery does not provide a sufficient level of detail. This information should enhance the forecasting capacity of species distribution models designed to predict  the evolution of vegetation as a result of climate warming.  

We present our methodology to achieve this goal after the envisaged approach had to be redesigned during the project, as the data collected by the world’s finest imaging spectrometer for the study did not meet the localization accuracy required. This highlights the importance of having the ability to implement and control all stages of the processing chain in-house and to present data openly in  their raw form. 

Presentation

This project enables precise mapping of elongated structures in inaccessible natural environment with small drones. Its implementation represents considerable savings for monitoring and management of man-made infrastructure and its protection in complex surroundings. The progress is possible through combination of state-of-the-art flying platform with advanced planning, on-line quality control of aerial vehicle position and accuracy prediction of 3D reconstruction

The objective of this project is to put into production a small, flexible and accurate airborne mapping system that integrates laser scanner, digital imagery and navigation sensors. Unique in its size and precision it can be embarked onto a helicopter within few minutes and can provide autonomous surface mapping of an area of interest with a high precision (0.2m) and resolution (<1m²), shortly after the flying mission. Its application span from natural hazards to corridors mapping.