Human Pilots Research Articles

Obstacle Avoidance Drone by Deep Reinforcement Learning and Its Racing with Human Pilot

Drones with obstacle avoidance capabilities have attracted much attention from researchers recently. They typically adopt either supervised learning or reinforcement learning (RL) for training their networks. The drawback of supervised learning is that labeling of the massive dataset is laborious and time-consuming, whereas RL aims to overcome such a problem by letting an agent learn with the data from its environment. The present study aims to utilize diverse RL within two categories: (1) discrete action space and (2) continuous action space. The former has the advantage in optimization for vision datasets, but such actions can lead to unnatural behavior. For the latter, we propose a U-net based segmentation model with an actor-critic network. Performance is compared between these RL algorithms with three different environments such as the woodland, block world, and the arena world, as well as racing with human pilots. Results suggest that our best continuous algorithm easily outperformed the discrete ones and yet was similar to an expert pilot.

The Prosthetic Impulse Revisited in <em>A.I. Artificial Intelligence</em>

The Prosthetic Impulse Revisited in <em>A.I. Artificial Intelligence</em>

An Air Combat Decision Learning System Based on a Brain-Like Cognitive Mechanism

The unmanned aerial vehicle (UAV) has emerged unexpectedly as a new force in the recent local wars. To occupy the high ground of military technology in the future, research on autonomous air combat using UAVs is extremely important. In this paper, we propose an intelligent air combat learning system inspired by the cognitive mechanism of the human brain. The air combat capability was divided into two parts: declarative and procedural memory. Imitating the updating and storage mechanism of human knowledge, a long- and short-term hierarchical asynchronous learning principle was designed. We adopted the basic idea of using the error signal to drive learning according to neurophysiological research. Drawing lessons from the working memory mechanism, the error signal for driving the short-term learning was created in the absence of labelled data, using only the interactive information. Then, we proved that the learning mechanism could ensure that system performance advances steadily and gradually. The experiments illustrated that the learning system designed in this paper can achieve some inferior confrontation ability through self-learning without human prior knowledge. Action strategies formed by learning are analogous to the classical tactical manoeuvres of human fighter pilots. We compared our work with related works and found that our method could improve its performance continuously and finally defeat its opponent.

Measurement of aircraft approach using airfield image

An aircraft approach to the runway is the most safety-critical phase of the flight. The measurement of aircraft position in the relation to the runway is crucial for proper flight control and proper decisions to control. There are several measurement methods used in aviation but human pilots in VFR conditions use images of airfield for the purpose of obtaining aircraft position. An airfield image and information about real location of markers in the runway coordinate frame are the basis for a new measurement method. After marker localisation, i.e. existing airfield objects and runway paintings, the aircraft position is computed by solving equations describing geometric relationships. Algorithm, measurement accuracy analysis and samples of in-flight experimental results are presented in this paper. Various factors, such as camera resolution, marker size and aircraft position are considered during the analysis, providing the possibility to recommend properties of the camera system. Although the existing airfield infrastructure can be directly used for measurements, further improvements are also proposed.

Charting the Edges of Human Performance

In the Horizon 2020 funded Future Sky Safety programme, the Human Performance Envelope project pushed airline pilots to the edges of their performance in real-time cockpit simulations, by increasing stress and workload, and decreasing situation awareness. The aim was to find out how such factors interact, and to detect the edges of human performance where some form of automation support should be employed to ensure safe continued flight. A battery of measures was used, from behavioural to physiological (e.g. heart rate, eye tracking and pupil dilation), to monitoring pilot performance in real time. Several measures – e.g. heart rate, heart rate variability, eye tracking, cognitive walkthrough, and Human Machine Interface (HMI) usability analysis – proved to be useful and relatively robust in detecting performance degradation, and determining where changes in information presentation are required to better support pilot performance in challenging situations. These results led to proposed changes in a prototype future cockpit human-machine interface, which were subsequently validated in a final simulation. The results also informed the development of a ‘Smart-Vest’ that can be worn by pilots to monitor a range of signals linked to performance.

Analysis of Human Interaction Patterns Emerging while Learning Agile Navigation of Unknown Environments

Abstract Trained human pilots or operators stand out through their efficient, robust, and versatile spatial guidance and control skills. The hypothesis is that trained operators learn sensory-motor primitives previously described as interaction patterns, and proposed as units of behavior for organization and planning of behavior. The interaction patterns emerge as a result of repeated interactions with a task environment. This paper extends a previously presented modeling and analysis framework based on interaction patterns to evaluate human learning of unknown environments. The paper presents a hierarchical clustering method to extract interaction patterns from trajectory data over successive runs, and uses these patterns for the analysis of the perceptual and control characteristics while accounting for different skill levels.

Attitude estimation using horizon detection in thermal images

The lack of redundant attitude sensors represents a considerable yet common vulnerability in many low-cost unmanned aerial vehicles. In addition to the use of attitude sensors, exploiting the horizon as a visual reference for attitude control is part of human pilots’ training. For this reason, and given the desirable properties of image sensors, quite a lot of research has been conducted proposing the use of vision sensors for horizon detection in order to obtain redundant attitude estimation onboard unmanned aerial vehicles. However, atmospheric and illumination conditions may hinder the operability of visible light image sensors, or even make their use impractical, such as during the night. Thermal infrared image sensors have a much wider range of operation conditions and their price has greatly decreased during the last years, becoming an alternative to visible spectrum sensors in certain operation scenarios. In this paper, two attitude estimation methods are proposed. The first method consists of a novel approach to estimate the line that best fits the horizon in a thermal image. The resulting line is then used to estimate the pitch and roll angles using an infinite horizon line model. The second method uses deep learning to predict attitude angles using raw pixel intensities from a thermal image. For this, a novel Convolutional Neural Network architecture has been trained using measurements from an inertial navigation system. Both methods presented are proven to be valid for redundant attitude estimation, providing RMS errors below 1.7° and running at up to 48 Hz, depending on the chosen method, the input image resolution and the available computational capabilities.

A Study of Percieved Stress Level among Engineers working in Government and Private Sector

Background: Stress is a natural concomitant of work life and is inevitable today. Stress has been examined among various groups of professionals such as accountants, managers, human service professionals, teachers, nurses and pilots, but little attention has been given to perceived stress level among engineers, and therefore present study was planned. Objective: ‘To assess the, ‘Perceived stress level among engineers working in government and private sector. Material and Methods: 60 engineers, 30 from private sector & 3o from government sector were taken from Mumbai (India). Among 30 engineers 15 were female engineers and 15 were male engineers, who were given Perceived stress scale (Cohen et al., 1988) along with semi structured interview. Results: indicated a) Perceived stress level among engineers working in private sector was more than perceived stress level among engineers working in government sector., b) Perceived stress level among female engineers was more than the perceived stress level in male engineers, c) Perceived stress level in female engineers working in private sector was more than perceived stress level in females working in government sector, d) Perceived stress level in male engineers working in private sector was more than perceived stress level in male engineers working in government sector. Conclusion: In nutshell, perceived stress level has a significant impact among male and female engineers working in private sector, as compared to engineers working in government sector. Therefore it is important to intervene.

Collision Detection for Underwater ROV Manipulator Systems.

Work-class ROVs equipped with robotic manipulators are extensively used for subsea intervention operations. Manipulators are teleoperated by human pilots relying on visual feedback from the worksite. Operating in a remote environment, with limited pilot perception and poor visibility, manipulator collisions which may cause significant damage are likely to happen. This paper presents a real-time collision detection algorithm for marine robotic manipulation. The proposed collision detection mechanism is developed, integrated into a commercial ROV manipulator control system, and successfully evaluated in simulations and experimental setup using a real industry standard underwater manipulator. The presented collision sensing solution has a potential to be a useful pilot assisting tool that can reduce the task load, operational time, and costs of subsea inspection, repair, and maintenance operations.

Benchmarking Brain-Computer Interfaces Outside the Laboratory: The Cybathlon 2016.

This paper presents a new approach to benchmarking brain-computer interfaces (BCIs) outside the lab. A computer game was created that mimics a real-world application of assistive BCIs, with the main outcome metric being the time needed to complete the game. This approach was used at the Cybathlon 2016, a competition for people with disabilities who use assistive technology to achieve tasks. The paper summarizes the technical challenges of BCIs, describes the design of the benchmarking game, then describes the rules for acceptable hardware, software and inclusion of human pilots in the BCI competition at the Cybathlon. The 11 participating teams, their approaches, and their results at the Cybathlon are presented. Though the benchmarking procedure has some limitations (for instance, we were unable to identify any factors that clearly contribute to BCI performance), it can be successfully used to analyze BCI performance in realistic, less structured conditions. In the future, the parameters of the benchmarking game could be modified to better mimic different applications (e.g., the need to use some commands more frequently than others). Furthermore, the Cybathlon has the potential to showcase such devices to the general public.

An anticipatory warning/advisory system for crew-aided flight status management

The aim of this paper is to sketch a model-based crew aided-decision system for anticipatory operational flight issues management. The system, termed Virtual co-Pilot (VcP), is intended to anticipate off-normal flight conditions and to support aircrew for better planning of upcoming flight phases and recovery from critical flight regimes. The core element is based on a distributed set observer and predictor which provides reachable regions for which safe operations of the aircraft can be guaranteed, and their projection over a specified time window which includes optimization of the prediction horizon. Based on available control authorities the system should also provide guidance and control options for remaining in a safe flight maneuvering envelope. The system will not interfere with the Flight Control System, and the human pilots will keep the overall responsibility.

Bull ants fine-tune ocelli to their lifestyle

Aeronautical engineers would love to know more about how flying insects control their manoeuvres. While human pilots are equipped with an artificial horizon for orientation, flying insects depend on three simple eyes (ocelli) set in the tops of their heads, which probably track the position of the horizon for stability. Intriguingly, some walking insects, such as members of the Myrmecia bull ant family, are also equipped with these unusual mini-eyes, even though they are probably of most use when members of a nest take to the wing to establish a new colony. Knowing that different members of the Myrmecia family, which often share the same territory, tend to be active at different times of day, Ajay Narendra wondered whether the insects have fine-tuned the specialised minute eyes to their different lifestyles.Armed with boots for protection from the ants’ fearsome stings, Narendra and Willi Ribi ventured out into the mountains and national parks in the vicinity of Canberra to collect workers of the walking day-active M. croslandi and dawn/day-active M. tarsata bull ants, while the dusk-active M. nigriceps and the nocturnal M. pyriformis ants had to be collected by infrared light at night, so that the pair could ‘spot them before they spotted us’. However, collecting flying members of each family was more challenging: ‘They fly out of the nest only once a year and were produced by only a few of the nests’, recalls Narendra, who eventually gathered night-flying M. nigriceps males and day-flying M. pyriformis males.Back in the lab, the pair took detailed images of the ants’ mini eye structures using light and electron microscopes. ‘The ocelli of the [walking] worker ants were surprisingly small compared to those of the flying ants’, says Narendra, who had to take great care handling the delicate structures. The day-active M. croslandi had the smallest ocelli lenses (76 μm), the species that were active in dim light and during the night had larger (129–201 μm) lenses, and the lenses of the flying ants were at least 10 times larger than those of the walking ants. The duo then compared the size of the light-sensitive portion of the eye, the rhabdom, across the four species of walking ants and found that the rhabdoms were widest in the ants that are active in dim-light conditions, probably to increase their light sensitivity, while the eyes of the flying ants were packed with 3 times as many rhabdoms as the eyes of the walking workers.Knowing that many insects are capable of sensing polarised skylight with their ocelli for navigation, Narendra and Ribi looked for structures in the ant ocelli that might detect light polarisation and were impressed to see that the two walking day-active species (M. croslandi and M. tarsata) had the structures, while the species that were active toward the dim end of the day lacked them. They also found that the ocelli of both species of flying ants were equipped with structures that may be sensitive to light polarisation, suggesting that the mini-eyes may be polarisation detectors, even though M. nigriceps flies at night.In addition, the two scientists was surprised when they discovered a reflective crystal structure in the ocelli of the flying night-active M. nigriceps males, toward the top of the rhabdom. Explaining that other animals use similar reflective eye coatings to enhance their sensitivity to light in dim conditions, Narendra suspects that the nocturnal fliers take advantage of the increased light sensitivity to make the most of what little light there is when they take to the wing.

An autonomous aerial combat framework for two-on-two engagements based on basic fighter maneuvers

Unmanned combat aerial vehicles are expected to assist human pilots or to perform autonomous aerial combat in the future. Previous research studies on autonomous aerial combat often have limitations due to overly simplifying assumptions, heavy computational loads, and limited scalability. In this paper, we propose a novel autonomous aerial combat algorithm with high-performance and real-time calculations based on basic fighter maneuvers for gun-based engagement within visual range. From the perspective of energy maneuver, which is the core concept of basic fighter maneuvers, we analyzed the flight envelope and maximum maneuvering point of a fighter jet model using the velocity-load factor and energy-maneuver diagrams. We designed a score function and matrix for situation assessment based on combat geometries, enabling cooperative maneuvers and role assignment. By performing gun modeling and target prediction based on the maneuvering plane, the impact point for gun shooting was obtained. The performance of the designed algorithm was validated through not only Monte-Carlo simulations with various initial conditions, but also engagements with another algorithm for comparison, and evaluation sessions by an Air Force combat instructor in a real-time simulation environment using virtual reality goggles and X-Plane.

No GPS, no pilot – no problem

Above ground, drones can navigate autonomously using GPS, but these satellite signals don't penetrate deep underground, meaning robot spelunkers normally require human pilots. Ahmed AlNomany and his colleagues at Swedish company Inkonova are working on an alternative. Giving the drone a view of its surroundings is the first step. Using laser scanners and a technique called SLAM, which calculates the distance between the drone and nearby objects, it will build up a map of the area around it. And it can do this fast. The company recently used its manually operated drone, TILT Ranger, to map an underground mine in Mali using SLAM. In just 10 minutes, the team was able to virtually reconstruct a section of the mine with a volume of roughly 30,000 cubic meters. Now the firm's new autonomous drone can combine its map with input from sensors, such as an accelerometer, to help position itself and move without GPS. So far, during preliminary tests, it has been able to stabilize itself on its own in the air

Mild Normobaric Hypoxia Exposure for Human-Autonomy System Testing

An experiment investigated the impact of normobaric hypoxia induction on aircraft pilot performance to specifically evaluate the use of hypoxia as a method to induce mild cognitive impairment to explore human-autonomous systems integration opportunities. Results of this exploratory study show that the effect of 15,000 feet simulated altitude did not induce cognitive deficits as indicated by performance on written, computer-based, or simulated flight tasks. However, the subjective data demonstrated increased effort by the human test subject pilots to maintain equivalent performance in a flight simulation task. This study represents current research intended to add to the current knowledge of performance decrement and pilot workload assessment to improve automation support and increase aviation safety.

Construct User Guide

Abstract : Construct is an agent-based network-centric simulation. Although frequently lumped together, agent-based simulations vary widely in complexity and computational cost some are extremely inexpensive (e.g., Swarm) and allow hundreds of thousands or even millions of agents to operate in the same simulation, while others are rather expensive and often require the support of an entire processor per agent (e.g., Soar or ACT-R). This increase in computational expense, however, is matched by construct validity to the actions of cognitively bounded humans: the least computationally expensive (per agent) simulations replicate the behavior of insects (specifically ants) while ACT-R has been able to replicate the brain activation patterns of children solving algebra problems and Soar has replicated fighter pilot operations in concert with human pilots.

Haptic feedback and visual servoing of teleoperated unmanned aerial vehicle for obstacle awareness and avoidance

Obstacle avoidance represents a fundamental challenge for unmanned aerial vehicle navigation. This is particularly relevant for low altitude flight, which is highly subjected to collisions, causing property damage or even compromise human safety. Autonomous navigation algorithms address this problem and are applied in various tasks. However, this approach is usually overshadowed by unreliable results in uncertain environments. In contrast, human pilots are able to maneuver vehicles in complex situations, in which an algorithm would no offer a reliable performance. This article explores a novel configuration of assisted flying and implements an experimental setup in order to prove its efficacy. The user controls an unmanned aerial vehicle with a force feedback device, where simultaneously an assisted navigation algorithm can manipulate this apparatus to divert the unmanned aerial vehicle from its path. Experiments confirm the authors’ hypothesis that the unmanned aerial vehicle is deviated or maintains the same course at the operator’s will. Unlike conventional controllers that dictate roll, pitch, and yaw, this implementation uses direct mapping between the position represented by the haptic device and the unmanned aerial vehicle. This configuration applies feedback before the unmanned aerial vehicle has reached the position referenced by the haptic device, providing valuable time for the user to make the necessary path correction.

Controlling a Fleet of Unmanned Aerial Vehicles to Collect Uncertain Information in a Threat Environment

Unmanned aerial vehicles (UAVs) have been proved to be successful and efficient for information collection in a modern battlefield, especially in areas that are considered to be dangerous for human pilots. Currently, a UAV is remotely controlled by a ground station through frequent data communications, which make the current system vulnerable in a threat environment. We propose a decentralized control strategy while requiring UAVs to maintain radio silence during the entire mission. The strategy is analyzed based on a scenario where a fleet of vehicles is assigned to search and collect uncertain information in a set of regions within a given mission time. We demonstrate that a region-sharing strategy is beneficial even when there is no extra reward gained from additional information collection. Implementing a region-sharing strategy requires solving a decentralized time allocation problem, which is computationally intractable. To overcome this, an approximate formulation is developed under an independence assumption for information collected by different vehicles. This approximate formulation allows us to decompose, by vehicle, the time allocation problem, and obtain an easily implementable policy that takes on a Markovian form. We develop a sufficient condition under which the approximate formulation becomes exact. A numerical study establishes the computational efficiency of the method; only a few CPU seconds are needed for problems with a planning horizon of 300 time units and 40 regions. We further present a case study to illustrate region-sharing behaviors among UAVs while using practical parameter values. Finally, we compare the obtained policy with the optimal policy found using a complete enumeration method for small instances. Under different parameter settings, the average optimality gap ranges from 0.23% to 19.90%. The e-companion is available at https://doi.org/10.1287/opre.2017.1590 .

Landing Site Reachability in a Forced Landing of Unmanned Aircraft in Wind

Autonomous contingency management systems, such as a forced-landing system, which reacts appropriately to an engine failure, are important for the safe operation of unmanned aircraft systems. This paper details a method to ascertain the reachability of any possible emergency landing site for a forced landing in steady uniform wind conditions. With knowledge of the aircraft’s state, such as speed, heading, location, and orientation of a landing site, a method to calculate a minimum height loss path is developed based on aircraft glide performance. Wind direction and speed are taken into account using a trochoidal approach by defining the minimum height loss turn path. To facilitate real-time implementation, simplified gliding equations are developed without accuracy loss. The reachability of each site can be calculated as well as how much safety margin an aircraft would have. This method is generic and could also provide decision support for human pilots in forced-landing situations. Two types of aircraft, the Airbus A320-400 and the Cessna 172, have been investigated to demonstrate the usefulness of the method, using Monte Carlo simulations in a synthetic X-Plane simulation environment, to demonstrate the performance and effectiveness of the proposed approaches.

Feature article: an advanced sense and collision avoidance strategy for unmanned aerial vehicles in landing phase

Unmanned Aerial Vehicles (UAVs) become promising in different civilian and military applications due to their lower cost and greater flexibility in comparison to manned aircraft. However, the growing diversity of flight makes UAVs vulnerable to mid-air collisions. To ensure the safety for manned aircraft, human pilots are responsible for detecting intruders in airspace and performing appropriate maneuvers to avoid collisions [1]. Unlike manned aircraft with human pilots involved, UAVs must be equipped with Sense and Avoid (SAA) systems to guarantee the flight safety [2]. Thus, SAA systems play an important role in merging UAVs into the National Airspace System. As illustrated in Figure 1, an SAA scheme is composed of four units, which are sensing, conflict detection, collision avoidance, and flight controller, respectively.