Project Dynamic VeloSLAM

Summary

Algorithms for solving the robotic simultaneous localization and mapping (SLAM) problem are a key scientific issue in mobile robotics research. Solutions to SLAM are of core importance in providing mobile systems with the ability to operate in true autonomy. SLAM algorithms integrate robot action and sensor readings and exploit the fact that previously mapped areas are recognized.

Popular mapping algorithms work with three degree of freedom (3 DoF) pose estimate. This choice is appropriate for indoor environments, but it is not sufficient for mapping many outdoor environments, which, in general, require using poses in 6 DoF in order to cope with elevation, pitch and roll. Therefore, the respective approaches are called 6D SLAM and consider all DoF of the pose of the mobile system with 3 position coordinates and roll, pitch and yaw angles. Core algorithms for solving 6D SLAM are part of the 3DTK – The 3D Toolkit which is an open-source project maintained by PI Nüchter, and which has been well-published.

The project "Dynamic VeloSLAM in Urban Environment" aims at making progress beyond the state of the art by dropping the constraint of mapping a static environment. Conventional robotic mapping approaches assume a static environment where changing elements, for example humans passing by, are simply treated as noise. On the contrary, dynamic mapping considers changes in the environment explicitly.

We will address the challenges in dynamic environments at different levels. First, as it is required in several robotic applications, we will investigate the ability to detect and describe changes with respect to a reference model of the environment. Second, detecting and removing dynamic parts in the current scan on the fly will allow for creating more accurate 3D models, for example in urban or in general populated environments (detection of non-stationary objects). In the same way, i.e., by disregarding non-stationary parts of a scan, this work will enable more accurate localization with respect to a given static model of the environment. Third, from a more conceptual view, dynamic mapping is the problem of continuous adaptation of maps over time. As such, it needs to address the stability-plasticity dilemma (the trade-off between adaptation to new patterns and preservation of old patterns). Of particular importance is that changes occur at different timescales and that they might be only temporary. Accordingly, a generic approach to dynamic mapping needs to represent these different types of changes properly together with the occluded part of the environment as it appeared before the change occurred. Further, new approaches to maintain and to use dynamic maps are required.

Funding

VeloSLAM is funded by the National Natural Science Foundation of China (NSFC), the China Scholar Council (CSC), and the German Academic Exchange Service DAAD.

Project period: 2012-2013

Partners

• Transportation Research Center, Wuhan University

  

• Automation group at Jacobs University gGmbH

  

Publications

• Ming Li, Wei Li, Jian Wang, Qingquan Li, and Andreas Nüchter Dynamic VeloSLAM – Preliminary Report on 3D Mapping of Dynamic Environments. In 2012 IEEE Intelligent Vehicles Symposium (IV '12), Workshop on Navigation, Perception, Accurate Positioning and Mapping for Intelligent Vehicles, Alcalá de Henares, Madrid, Spain, 2012 [Get Paper (PDF)] [Talk (PDF)].

• Ming Li, Wei Li, Jian Wang, Qingquan Li, and Andreas Nüchter Towards Reliable Object Anchoring in Highly Dynamic Traffic Scenes in ICRA 2012 WORKSHOP Semantic Perception and Mapping for Knowledge-enabled Service Robotics (with interactive session and demonstrations), St. Paul, MN, USA, May 2012.