Long-term Autonomous Navigation

Research Vision

In this line of work, we are conducting research on autonomous operation over longer times. The key questions to address is how to efficiently represent the environment, how to update the map (environmental model) and how to encode and extract information regarding long term aspects, for example, the level of dynamics in the different regions in the map.

In this work both camera images as well as 2D / 3D laser data are used.

Notable Results

Long-term mapping of industrial environments

We have utilized the 3D-NDT-OM framework to work with handling long term data, for example, 24 hours continuous operation in a diary production unit (top movie). The 3D-NDT-OM handles dynamics and can also encodes the different levels of dynamics (bottom video). One utilization of different dynamic levels is, for example, to considered areas which are static to be more trustworthy for localization.

High precision localization in dynamic environments

As a usage of the NDT based maps a Mote-Carlo based and UKF based localization system has been integrated with good performance.

The localization system relying on the NDT based representation has been integrated into Kollmorgen’s navigation solutions called ‘Natural navigation‘.

Visual place recognition

To allow for autonomous system to utilize cameras, which are cost efficient as well as very information, as the main source of information there are multiple challenges to be addressed. The variation in images over time, such as light changes and shadows, different weather conditions and seasons have very large impact on the visual content. One results is a survey paper that covers different approaches for appearance based place recognition using image data.