Sense and Act

ARCAA research in Sense and Act, refers to the process in which a UAS is able to detect (sense) non-cooperative traffic and perform avoidance manoeuvres (Act) where necessary. This capabilty is being recognized as one of the key enablers for future UAS applications and integration in non-segregated airspace.

The Sense-and-Act program emcompases several topics such as:

1. Computer Vision based “Sense and Act” for UAS
2. UAV Guidance & Control for UAVs “Sense and Act”
3. Detection of Dim Sub-pixel Targets for UAV Collision Avoidance using the Hidden Markov Model and Morphologically Processed Imagery

1. Computer Vision based “Sense and Act” for UAS

This research focuses on developing algorithms to provide an Electro-Optic (EO) based sense and act capability suitable for UAS applications in daytime VFR conditions. The use of visual-spectrum digital camera sensors offers a relatively low-cost, low weight and low power alternative to radar-based solutions. A high level overview of the approach is outlined below.

Video of a near miss between a Lunar UAS and an Ariana Afghan Airlines Airbus A300B4 (Kabul, August 2004)

Description of Research

An initial phase is required to reduce an incoming image containing millions of pixels down to a handful of candidate target features for future processing. As a result of the geometries involved in a typical air-to-air collision scenario, a collision-course aircraft will appear as a small, point-like target exhibiting little motion in the image plane. The following techniques are used:

A subsequent, more computationally intensive, phase in then required to distinguish between genuine collision threats and residual target features caused by point like clutter (e.g. a distant house). Some of the properties examined as part of this process include:

Another focus of this research is to objectively compare performance results against the measured performance of a human observer, with the intention of demonstrating to regulatory bodies that the current benchmark of “equivalent human performance” can be met. An example of such comparison, taken from real-world data, is displayed below.

Current results of our Sense-and-Act algorithm. Algorithm detects aircraft 14.5 seconds before human observer is able to see the vehicle.

Much of this work is currently being practically implemented in the Smart Skies Project.

Publications

• Vision-based Sense-and-Act Publications (QUT ePrints)

Related Sites

• Sense-and-Act Research Theme
• Smart Skies Project

Research Personnel

• Mr Ryan Carnie
• Dr. Luis Mejias

Related Media

2. UAV guidance and Control

This research deals with methods of guidance and control that might be applied to an autonomous aerial vehicle to guarantee a successful uncooperative collision scenario mitigation. 

Video of a near miss between a Lunar UAS and an Ariana Afghan Airlines Airbus A300B4 (Kabul, August 2004)

Description of Research

This research is based on information input from a vision-based sensor coming under the classical "See and Avoid" banner.  Thus the research actually may have relevance to manned aviation as well.

This research is motivated since many professional bodies like RTCA, ASTRAEA etc have identified Collision Avoidance as the highest-priority technical barrier to UAV airspace integration.

A recent investigation into current high-performance Guidance methods is being performed.  At the leading edge of these is Lypunov Vector Fields (LVF) and results are shown here for globally stable trajectories generated for attraction towards a limit cycle defined by the Hopf bifurcation.

Commands generated by the algorithm, where red circle is the target trajectory

Publications

• UAV Guidance and Control Publications (QUT ePrints)

Related Sites

• Sense-and-Act Research Theme
• Smart Skies Project

Research Personnel

• Mr Shane Degen
• Dr. Luis Mejias
• Dr. Jason Ford

Related Media

3. Detection of Dim Sub-pixel Targets

Collision threats often appear as small, dim, constant bearing 'spots' - this research focuses on the development of algorithms for the detection of targets exhibiting these properties within images.

Description of Research

Unmanned aerial vehicles (UAVs) represent a unique class of aircraft where there are no human pilots onboard.  Until recently, the use of UAVs has been almost exclusive to the military.  In just a relatively short period of time spanning the past few years, the push for commercial/civil applications of UAVs has gathered momentum at an incredible pace.  So rapid has support and interest for commercial/civil UAVs grown in recent times that many would consider the technology to be maturing at a rate far outstripping the development of necessary legislative and regulatory frameworks.

In the past when UAVs were by and large constrained to the military, aviation authorities around the world adopted a policy of 'segregation', whereby manned and unmanned aircraft were kept apart in distinctly separate airspaces.  This policy of segregation has now been widely acknowledged as being unworkable and too restrictive for commercial/civil UAVs.  The notion of an inevitable shift towards 'integration', whereby manned and unmanned aircraft are permitted to share the same airspace, is starting to gain widespread acceptance among those in the aviation industry.

To have UAVs and manned aircraft sharing the same airspace on a regular basis is unprecedented.  There are numerous issues that need to be resolved before this can occur safely and effectively.  They include the handling of increased air traffic volumes, management of communication spectrums, and addressing public concerns/perceptions.  However, arguably the issue of greatest concern to aviation authorities is that of collision avoidance or 'sense-and-avoid' - i.e. how to ensure safe separation between manned aircraft and UAVs, as well as between UAVs themselves.

The ability to detect potential collision-course threats in the sky is a key element in the overall sense-and-avoid concept.  To date, there is an abundance of algorithms published in literature that have been designed for (or may be applicable to) the detection of airborne objects.  In particular, continued advances in image processing and other related fields have made computer-vision techniques more and more attractive.

Computer-vision techniques based around a dynamic programming algorithm have been identified in the literature as a leading candidate for the detection of airborne obstacles.  The purpose of this research is to propose an alternative computer-vision technique based on the hidden Markov model (HMM) for detecting slow-moving point-like targets that are likely to feature in a collision avoidance scenario. The algorithm is based on the diagram below.. 

Illustration of the target detection algorithm

A key advantage of the HMM is that it has a well developed statistical formalism which enables appropriate parameters to be readily varied and accounted for. A visualisation of the outputs of the HMM approach are shown below.

Illustration of HMM process for point-like target detection in images

Publications

• Detection of Dim Sub-pixel Targets Publications (QUT ePrints)

Related Sites

• Sense-and-Act Research Theme
• Smart Skies Project

Research Personnel

• Mr John Lai
• Dr. Jason Ford
• Prof. Peter O'Shea

Related Media