Unmanned Ground Vehicle Center (UGVC)

Develop to be mobile

Research Topics

Visual 3D Perception

Behavioral Based Navigation

SLAM

Modern intelligent robots are generally equipped with an abundance of sensors like for example GPS, Laser, ultrasound sensors, etc to be able to navigate in an environment. However, this stands in contrast to the ultimate biological example for these robots: us humans. Indeed, humans seem perfectly capable to navigate in a complex, dynamic environment using only vision as a sensing modality. This observation inspired us to investigate visually guided intelligent mobile robots.

In order to understand and reason about its environment, an intelligent robot needs to be aware of the three-dimensional status of this environment. The problem with vision, though, is that one only receives a two-dimensional image as input. Recovering 3D-information has been in the focus of attention of the computer vision community for a few decades now, yet no all-satisfying method has been found so far. Most attention in this area has been on stereo-vision based methods, which use the displacement of objects in two (or more) images. Where stereo vision must be seen as a spatial integration of multiple viewpoints to recover depth, it is also possible to perform a temporal integration. The problem arising in this situation is known as the "Structure from Motion" problem and deals with extracting 3-dimensional information about the environment from the motion of its projection onto a two-dimensional surface.

Cognitive research has shown that the human brain uses no less than 12 different cues to estimate depth. If we want to achieve the same performance as the human vision system, we should therefore probably need to integrate all these different cues. In order to advance research in this field, we investigate at RMA the possibilities of stereo vision as well as Structure from Motion.

In order to localize itself and move toward its target, a mobile robot needs a description or a model of its working environment. This model is not always available and hence the robot should have the means to build such a model over time. This problem is known as mapping problem.

Mapping and localization are interlinked problems: If the robot's pose (spatial position and orientation) is known all along, building a map would be quite simple. Conversely, if a map of the environment exists already, it will be very easy to determine accurately the robot's pose at any time. In combination, however, the problem is much harder. The literature refers to the mapping problem often in conjunction with the localization problem named as Simultaneous Localization And Mapping (SLAM).

In our application the robot uses a single monocular camera to extract natural features in the scene. These features are used as landmarks in the built map. The proposed method builds several size limited maps based on the extracted visual features; combines them in a global map using an 'history memory' which accumulates sensory evidence over time to identify places with a stochastic model of the correlation between features; and finally, integrates data from GPS to localize the built map and therefore the robot in geo-referenced images in order to allow robot localization in a user defined global coordinate frame.

A control architecture describes the strategy to combine the three main capabilities of an intelligent mobile robot: sensing, reasoning (intelligence) and actuation. These three capabilities have to be integrated in a coherent framework in order for the mobile agent to perform one or more tasks adequately.

In behavior-based architectures, the reasoning or intelligence is incorporated by multiple behaviors. Based on selective sensory information, each behavior produces immediate reactions to control the robot with respect to a particular objective, a narrow aspect of the robot’s overall task such as obstacle avoidance or wall following. Behaviors with different and possibly incommensurable objectives may produce conflicting actions that are seemingly irreconcilable. Thus a major issue in the design of behavior-based control systems is the formulation of effective mechanisms for coordination of the behaviors’ activities into strategies for rational and coherent behavior. This is known as behavior fusion problem.

At RMA, we investigate new theorems of solving the behavior development and fusion problem. Classical behavior-based approaches generally do not allow for input coming from time-consuming modeling procedures such as Simultaneous Localization and Mapping (SLAM) or 3D Reconstruction. We aim to develop a theorem to incorporate this information in a behavior-based framework. Moreover, we develop new behavior fusion techniques to better handle the uncertainty present in the sensor measurements.

3D Simulation of Multi-Robot Systems

Before testing the robots in real environments, simulation can offer many advantages. It is far more cost-effective than real robots and sensors, particularly when experimenting with multi-robot systems. And the cost of failure is practically negligible with simulations: the robots cannot suffer physical damage in a virtual experiment. This increases safety when developing and testing new control applications and thus more risky situations can be explored.

The effort to test behaviors in the real world is significantly higher than an equivalent experiment in a simulated environment; therefore simulation allows focusing on intelligence and control.

Moreover, deterministic simulations are possible which allow to compare different algorithms independent of variations in the test set-up and to refine control strategies. Development time can thus be reduced by trying different scenarios and algorithms before experimenting in a real environment.

Although simulation provides many other advantages it’s not a substitute for real world experiments which are essential. By combining respectively the advantages of virtual and real world experiments efficiency can be improved.

At the RMA the goal is to develop a real-time 3D multi-robot simulator that fulfills a number of requirements: realistic physics simulation, collision detection and appropriate reaction, networking, CORBA interfaces, user interaction with on-line simulation, sensor and actuator simulation, etc.