Doctoral Research
Over the past decade, multirotor vehicles have evolved from popular research subjects to mature robotic platforms that are entering commercial use beyond recreational or experimental applications. As multirotor vehicles perform an increasingly large set of tasks, the need to increase their robustness and reliability and extend their capabilities emerges. Dario’s research aims at increasing the maneuverability of multirotor vehicles and is carried out in the Flying Machine Arena, an aerial vehicle test bed at ETH Zürich.
Quadrocopter Pole Acrobatics
[expand title=”show more…” swaptitle=” ” id=”pole-acrobatics” excerpt=”Through a combination of nonlinear, optimal control strategies and learning algorithms, the flight performance of quadrocopters can be dramatically improved, enabling the execution of complex, accurate, high-performance tasks. As an example, we have developed a system that allows quadrocopters to balance an inverted pendulum, throw it into the air, and catch and balance it again on a second vehicle.” excerptpos=”above-trigger” swapexcerpt=””]
Quadrocopters are among the most popular and widespread multirotor vehicles because of their mechanical simplicity and high maneuverability. Their inherently unstable and nonlinear dynamics pose interesting challenges for controller design and motivate research into nonlinear control strategies. Furthermore, although their dynamics are relatively straightforward to model in steady conditions, complex and difficult to model aerodynamic effects act on the vehicle when maneuvering at high speeds. These difficult to model effects diminish the effectiveness of model-based control approaches and instigate the development of learning strategies to compensate for unmodelled dynamics.
Through a combination of nonlinear, optimal control strategies and learning algorithms, the flight performance of quadrocopters can be dramatically improved, enabling the execution of complex, accurate, high-performance tasks. As an example, we have developed a system that allows quadrocopters to balance an inverted pendulum, throw it into the air, and catch and balance it again on a second vehicle. More details can be found in this research paper.
show less …
[/expand]
Attitude Control for Multirotor Vehicles
[expand title=”show more…” swaptitle=” ” id=”attitude-control” excerpt=”We have developed an attitude tracking control strategy and control allocation method that prioritizes the alignment of the vehicle’s crucial thrust direction if the available control effort becomes scarce, yielding improved performance when tracking aggressive maneuvers.” excerptpos=”above-trigger” swapexcerpt=””]
Traditional multirotor vehicles are under-actuated, i.e. unable to control all of their six degrees of freedom independently. In order to maximize performance criteria such as flight duration, range or payload, all rotors are typically arranged in a single plane, and as a consequence, the total thrust is limited to a single direction normal to the rotor disks. The position and attitude dynamics of multirotor vehicles are thus coupled and the direction of acceleration is determined by the vehicle’s attitude. A crucial requirement for the successful control of a multirotor vehicle is therefore the ability to accurately control the vehicle’s attitude. However, only two of the attitude’s three degrees of freedom are relevant of the position dynamics; the alignment of the vehicle about its thrust direction does not affects its position dynamics.
Driven by the increasing use of multirotor vehicles for various applications, multirotor vehicles are expected to encounter and recover from an increasingly large set of potential disturbances and perform increasingly aggressive flight maneuvers. However, even today’s state-of-the-art control strategies often struggle when recovering from large disturbances or when tracking aggressive flight maneuvers. In such scenarios, the commanded rotor thrusts can often not be generated due to saturation limits of the individual rotor thrusts. Typically, the rotor thrust commands are naively clipped at their saturation limits. However, this can significantly degrade the performance and may even lead to instabilities. Furthermore, clipping the rotor thrust commands does not account for the different importance of the vehicle’s total thrust and torques for the vehicle’s flight performance and stability. As an example, the position dynamics are invariant to the orientation of the vehicle about its thrust direction, thus rendering torques about the vehicle’s thrust direction less important than the vehicle’s total thrust or torques perpendicular to the thrust direction. We have therefore developed an attitude control and control allocation strategy for multirotor vehicles that prioritizes the alignment of the vehicle’s thrust direction if the available control effort becomes scarce in order to improve its trajectory tracking and disturbance recovery performance. More details can be found in this research paper.
show less …
[/expand]
An Omni-Directional Multirotor Vehicle
[expand title=”show more…” swaptitle=” ” id=”omnicopter” excerpt=”We have developed a fully-actuated multirotor vehicle that allows for the generation of thrust and torque in any direction and thereby decouples the position and attitude dynamics. In particular, we have developed an omni-directional multirotor vehicle, i.e. a vehicle capable of generating sufficient thrust to hover at any attitude and accelerate in any direction.” excerptpos=”above-trigger” swapexcerpt=””]
Early multirotor vehicle applications such as aerial photography or filming made use of the hover capability and unique aerial perspective provided by multirotor vehicles. Motivated by the rapid progress in multirotor vehicle technology and their ability to move freely in space, overcome obstacles on the ground, and reach otherwise inaccessible locations, an increasing number of researchers started to investigate the use of multirotor vehicles for other tasks including object manipulation, aerial construction, interaction with humans, and physical inspection. In these tasks, multirotor vehicles are physically interacting with the environment. However, the inability of traditional multirotor vehicles to independently generate thrusts and torques in any direction limits their suitability for tasks where there is physical interaction, as these often require the ability to suppress arbitrary force and torque disturbances instantaneously. Furthermore, due to the coupling of the vehicles’ position and attitude dynamics, the set of feasible position and attitude maneuvers is severly limited. To cope with the constrained motion range of traditional multirotor vehicles in aerial manipulation tasks, often robot arms with multiple degrees of freedom are attached to the vehicles, which yields systems that are very complex to handle.
We have developed a fully-actuated multirotor vehicle that allows for the generation of thrust and torque in any direction and thereby decouples the position and attitude dynamics. In particular, we have developed an omni-directional multirotor vehicle, i.e. a vehicle capable of generating sufficient thrust to hover at any attitude and accelerate in any direction. To fully exploit the vehicle’s unconstrained range of motion, a control strategy has been devised that enables the vehicle to simultaneously track a position and attitude trajectory. More details can be found in this research paper.
show less …
[/expand]
Computationally Efficient Trajectory Generation for Fully-Actuated Multirotor Vehicles
[expand title=”show more…” swaptitle=” ” id=”trajectory-generation” excerpt=”We have developed a method to generate trajectories that take full advantage of the dynamic capabilities of fully-actuated multirotor vehicles. The method is based on computational lightweight motion primitives and allows for the generation of several hundreds of thousands of trajectories per second. This enables the method to be embedded in a high level path planner that computes a large set of trajectories that achieve a desired high level goal, for example catching a ball.” excerptpos=”above-trigger” swapexcerpt=””]
A key feature required for the autonomous operation of a multirotor vehicle is the ability to plan trajectories that guide the vehicle from any desired initial state to any desired target state, while respecting the vehicle’s dynamics and input constraints. Ideally, a trajectory generator is capable of taking full advantage of the vehicle’s dynamic capabilities and is able to compute trajectories in real time, such that it can deployed in dynamic environments where the target state may change or obstacles may move in mid-flight.
We have developed a method to generate trajectories that take full advantage of the dynamic capabilities of fully-actuated multirotor vehicles. The method is based on computational lightweight motion primitives and allows for the generation of several hundreds of thousands of trajectories per second. This enables the method to be embedded in a high level path planner that computes a large set of trajectories that achieve a desired high level goal, for example catching a ball (see video below). Moreover, it allows the trajectory generator to be used as an implicit feedback law be regenerating the trajectory at each controller update step, and applying the initial inputs of the trajectory similarly to model predictive control. More details can be found in this research paper.
show less …
[/expand]
Other Research Projects
A Delay-Compensated Variometer
[expand title=”show more…” swaptitle=” ” id=”variometer” excerpt=”In order to find thermals that allow the ascent to high altitudes, paraglider pilots often use variometers. However, traditional variometers suffer from delays, which makes it hard to accurately thermals. In order to simplify this, Koni Schafroth in collaboration with Dario have developed a novel, delay-free variometer, the XC-Tracer.” excerptpos=”above-trigger” swapexcerpt=””]
Paragliders typically rely on variometers, i.e. climbing rate indicators, to find thermals that allow them to climb to high altitudes and cover large distances. Traditional variometers use barometers to measure the altitude and use the difference between two subsequent measurements to compute the climbing rate. Since the computation of the climbing rate requires two measurements, the climbing rate cannot be computed instantaneously and is only available with a delay. Furthermore, in order to be less prone to measurement errors, the computed climbing rate needs to be low-pass filtered, which introduces further delays. Due to the high speeds at which paraglider fly, even small delays in the climbing rate can make it very difficult for paraglider pilots to accurately locate thermals and identify their core where their uplift is the largest.
In order to simplify the process of finding thermals, Koni Schafroth developed a novel type of variometer, the XC-Tracer. In addition to a barometer, the XC-Tracer is also equipped with an inertial measurement unit, which provides acceleration and angular velocity data, and a GPS unit, which provides position data. Dario helped Koni with bringing state-of-the-art state estimation techniques to the XC-Tracer to fuse the barometer, inertial measurement unit and GPS data to an accurate and delay-free climbing rate estimate. The delay-free climbing rate estimate of the XC-Tracer allows paragliders to effectively identify thermals and is used by various word-class paragliders in competitions.
show less …
[/expand]
User Intent Detection for Walking Assistance Robots
[expand title=”show more…” swaptitle=” ” id=”exoskeleton” excerpt=”Exoskeleton robots are increasingly used as mobility aids for paraplegics as they offer many advantages compared to conventional devices such as wheelchairs. By supporting paraplegics while walking in a standing posture with their own legs and taking part in ordinary daily activities, exoskeleton robots can enhance a paraplegic’s life quality and self-esteem. We have developed a method that allows paraplegics to intuitively control the walking parameters of an exoskeleton robot such as, for example, stridge length and walking direction.” excerptpos=”above-trigger” swapexcerpt=””]
Several wearable exoskeleton robots have been developed in the past years that assist paraplegics in walking. A key component of these walking support systems is their user interface. The interface has to be intuitive to use and should be able to determine the user’s walking intention, i.e. the desired stride length, height difference between the subsequent strides, walking direction, and operation mode, for example, walking or climbing stairs. Only if the exoskeleton robot’s motion is synchronized with the patient’s intention, then the patient feels comfortable using the exoskeleton robot. Early research into exoskeleton robots aimed at enhancing the capabilities of healthy humans in order to, for example, carry heavier weights, run faster or walk longer distances. In these systems, the user’s walking intention was often derived from biological signals, for example, by measuring the myoelectric signals of the wearer’s lower body. However, such an approach is not feasible for paraplegics since these biological signals do not reach their lower body.
We have developed a method that allows paraplegics to intuitively control the exoskeleton’s motion using an inertial measurement unit mounted on crutches. The feasibility of the approach has been demonstrated in a trial where paraplegics in a rehabilitation center used the method to control an exoskeleton that enabled them to walk. More details can be found in this research paper.
show less …
[/expand]