Deep Drone Acrobatics

Kaufmann, Elia; Loquercio, Antonio; Ranftl, René; Müller, Matthias; Koltun, Vladlen; Scaramuzza, Davide

doi:10.15607/rss.2020.xvi.040

Cited by 123 publications

(124 citation statements)

References 25 publications

Supporting

Mentioning

110

Contrasting

Order By: Relevance

“…The realworld experiments confirm the findings obtained in simulation, i.e. visual information is crucial to compensate the drift of inertial measurements during long maneuvers [Kaufmann et al, 2020]. We refer the reader to the supplementary video to understand the dynamic nature of the experiments.…”

Section: Abstraction Helps Training and Generalizationsupporting

confidence: 81%

“…This intermediate representation is more consistent across simulation and reality than raw visual input. We formally show that training a network on abstraction of sensory input reduces the gap between simulation and reality [Kaufmann et al, 2020].…”

Section: Training Methodologymentioning

confidence: 99%

“…The network is trained with an off-policy learning approach. More details about the training procedure can be found in the main manuscript [Kaufmann et al, 2020]. Figure 3: Tracking error (left) and success rate (right) over time when a maneuver is executed repeatedly in simulation.…”

Section: Deep Sensorimotor Controllermentioning

confidence: 99%

See 2 more Smart Citations

Deep Drone Acrobatics (Extended Abstract)

Kaufmann

Loquercio

Ranftl

et al. 2021

Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence

Self Cite

View full text Add to dashboard Cite

Acrobatic flight with quadrotors is extremely challenging. Maneuvers such as the loop, matty flip, or barrel roll require high thrust and extreme angular accelerations that push the platform to its limits. Human drone pilots require years of practice to safely master such maneuvers. Yet, a tiny mistake could make the platform lose control, and brutally crash. This short paper describes an approach to safely train acrobatic controllers in simulation and deploy them with no fine-tuning zero-shot transfer on physical quadrotors. The approach uses only onboard sensing and computation.

show abstract

Section: Abstraction Helps Training and Generalizationsupporting

confidence: 81%

Section: Training Methodologymentioning

confidence: 99%

See 1 more Smart Citation

Deep Drone Acrobatics (Extended Abstract)

Kaufmann

Loquercio

Ranftl

et al. 2021

Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence

Self Cite

View full text Add to dashboard Cite

show abstract

“…Very little work exists on agile control of quadrotors at speeds beyond 5 m s −1 and accelerations above 2g, [1][2][3][4][5][6][7][8]. Even though these works show agile control at various levels, none of them accounts for aerodynamic effects.…”

mentioning

confidence: 99%

Data-Driven MPC for Quadrotors

Torrente

Kaufmann

Fohn

et al. 2021

IEEE Robot. Autom. Lett.

Self Cite

171

139

View full text Add to dashboard Cite

Aerodynamic forces render accurate high-speed trajectory tracking with quadrotors extremely challenging. These complex aerodynamic effects become a significant disturbance at high speeds, introducing large positional tracking errors, and are extremely difficult to model. To fly at high speeds, feedback control must be able to account for these aerodynamic effects in real-time. This necessitates a modeling procedure that is both accurate and efficient to evaluate. Therefore, we present an approach to model aerodynamic effects using Gaussian Processes, which we incorporate into a Model Predictive Controller to achieve efficient and precise real-time feedback control, leading to up to 70% reduction in trajectory tracking error at high speeds. We verify our method by extensive comparison to a state-of-theart linear drag model in synthetic and real-world experiments at speeds of up to 14m/s and accelerations beyond 4g. SUPPLEMENTARY MATERIALVideo: https://youtu.be/FHvDghUUQtc Code: https://github.com/uzh-rpg/data driven mpc I. INTRODUCTIONAccurate trajectory tracking with quadrotors in high-speed and high-acceleration regimes is still a challenging research problem. While autonomous quadrotors have seen a significant gain in popularity and have been applied in a variety of industries ranging from agriculture to transport, security, infrastructure, entertainment, and search and rescue, they still do not exploit their full maneuverability. The ability to precisely control drones during fast and highly agile maneuvers would allow to not only fly fast in known-free environments, but also close to obstacles, humans, or through openings, where already small deviations from the reference have catastrophic consequences.Operating a quadrotor at high speeds and controlling it through agile, high-acceleration maneuvers requires to account for complex aerodynamic effects acting on the platform. These effects are difficult to model, since they consist of a combination of propeller lift and drag dependent on the induced airstream velocity, fuselage drag, and complex or even turbulent effects due to the interaction between the propellers, the downwash of other propellers, and the fuselage.

show abstract

“…Comparisons with different techniques of RL and Proportional-Integral-Derivative (PID) controllers are presented. Similarly, Kaufmann et al [28] propose an intelligent control system for quadrotors to perform autonomous flights with extreme acrobatic maneuver, using only onboard sensing and computation. The system training uses demonstrations from an optimal controller with privileged feedback information in simulations.…”

Section: Controlmentioning

confidence: 99%

An integrated solution for an autonomous drone racing in indoor environments

Rezende

Miranda

Rezeck

et al. 2021

Intel Serv Robotics

View full text Add to dashboard Cite

The use of drones is becoming more present in modern daily life. One of the most challenging tasks associated with these vehicles is the development of perception and autonomous navigation systems. Competitions such as Artificial Intelligence Robotic Racing (AIRR) and Autonomous Drone Racing were launched to drive the advances in such strategies, requiring the development of integrated systems for autonomous navigation in a racing environment. In this context, this paper presents an improved integrated solution for autonomous drone racing, which focuses on simple, robust, and computationally efficient techniques to enable the application in embedded hardware. The strategy is divided into four modules: (i) A trajectory planner computes a path that passes through the sequence of desired gates; (ii) a perception system that obtains the global pose of the vehicle by using an onboard camera; (iii) a localization system which merges several sensed information to estimate the drone's states; and, (iv) an artificial vector field-based controller in charge of following the plan by using the estimated states. To evaluate the performance of the entire integrated system, we extended the FlightGoggles simulator to use gates similar to those used in the AIRR competition. A computational cost analysis demonstrates the high performance of the proposed perception method running on limited hardware commonly embedded in aerial robots. In addition, we evaluate the localization system by comparing it with ground truth and when there is a disturbance in the location of the gates. Results in a representative race circuit based on the AIRR competition showed the proposed strategy's competing performance, presenting itself as a feasible solution for drone racing systems.

show abstract

Deep Drone Acrobatics

Cited by 123 publications

References 25 publications

Deep Drone Acrobatics (Extended Abstract)

Deep Drone Acrobatics (Extended Abstract)

Data-Driven MPC for Quadrotors

An integrated solution for an autonomous drone racing in indoor environments

Contact Info

Product

Resources

About