MBZIRC 2019 - Proposal
-
A summary of achievements obtained during participation of team of CTU in Prague, UPENN and UOL in MBZIRC 2017. The entire team was led by PI of the proposal Martin Saska and obtained 1st place in the 3rd challenge, 2nd place in the 1st challenge, and 3rd place in the Grand Challenge out of 450 teams that indicated interest to participate (143 applicants from the most prestigious universities in the world was considered for participation and 24 of them selected for final rounds in Abu Dhabi).
Fastest fully autonomous landing on a moving car during MBZIRC 2017 competition.The team of CTU in Prague, UPENN and UOL, led by PI Martin Saska, achieved the fastest time of fully autonomous landing on a moving car in the entire MBZIRC 2017 competition. |
System verification in challenging desert environment prior MBZIRC 2017 competition.Deployment of the system designed for the MBZIRC competition in challenging desert environment. A team of 3 MAVs were able to cooperatively solve the pick and place task fully autonomously (the system was fully autonomous from take-off to landing) and reliably (without any accident during numerous repeated trials) in strong wind conditions. |
Hunting an intruder drone with autonomous UAV.Verification of usefulness of the approach taking advantage of predictive nature of the tracker, where MAV conducts proper feedforward maneuvers while flying outside of typical operation point of linearity, thanks to the used non-linear SO(3), for interaction with intruder UAV. A balloon was attached on a randomly flying drone and the system was able to repeatedly hit the balloon by a fully autonomous MAV - only a control system was tested and both MAVs were localised by RTK GPS. |
Cooperative Transportation using Small Quadrotors using Monocular Vision and Inertial Sensing.This video presents the state estimation, control, and trajectory planning in cooperative transportation of structures, which are either too heavy or too big to be carried by small micro vehicles. We consider small quadrotors, each equipped only with a single camera and IMU as a sensor. We present a new approach to coordinated control which allows independent control of each vehicle while guaranteeing the system’s stability and a new cooperative localization scheme that allows each vehicle to benefit from measurements acquired by other vehicles. The later is formulated as an optimization problem exploiting the rigid structure information. We reach speeds of over 4.2 m/s, which angular rates close to 300 deg/s and accelerations up to 5 m/s^2. |
Aggressive Flight with Suspended Payloads using Vision-based Control.Payload manipulation with autonomous aerial vehicles has been an active research area for many years. In particular, recent approaches have sought to plan, control, and execute maneuvers with large, yet deliberate, load swings for more agile, energy-optimal maneuvering. Unfortunately, the system's fast, nonlinear dynamics makes executing such trajectories a significant challenge and experimental demonstrations thus far have relied on nontrivial simplifications. Rather than relying on a motion capture system, we estimate the state of the payload using a downward facing monocular camera and Inertial Measurement Unit (IMU) on board a quadrotor. We demonstrate closed-loop feedback control of the payload position in the full three-dimensional workspace and execute a complete planning, estimation, and control pipeline on an onboard processor. We demonstrate robust payload control across different system and task parameters with payload swings of up to 50 degrees from the vertical axis. To the best of our knowledge, this represents the first demonstration of agile maneuvers with closed-loop payload control and the largest payload angle achieved in experiments. |
Robust Stereo Visual Inertial Odometry for Fast Autonomous Flight.In recent years, vision-aided inertial odometry for state estimation has matured significantly. However, we still encounter challenges in terms of improving the computational efficiency and robustness of the underlying algorithms for ap- plications in autonomous flight with micro aerial vehicles in which it is difficult to use high quality sensors and powerful processors because of constraints on size and weight. In this paper, we present a filter-based stereo visual inertial odometry that uses the Multi-State Constraint Kalman Filter (MSCKF) [1]. Previous work on stereo visual inertial odometry has resulted in solutions that are computationally expensive. We demon- strate that our Stereo Multi-State Constraint Kalman Filter (S-MSCKF) is comparable to state-of-art monocular solutions in terms of computational cost, while providing significantly greater robustness. We evaluate our S-MSCKF algorithm and compare it with state-of-art methods including OKVIS, ROVIO, and VINS-MONO on both the EuRoC dataset, and our own experimental datasets demonstrating fast autonomous flight with maximum speed of 17.5m/s in indoor and outdoor environments. |
Interior documentation in a church in Šternberk.Documentation of the interior in a church in Šternberk by a pair of autonomous MAVs. The self-stabilized MAVs are used for filming locations with insufficient illumination. A leader MAV carries a camera while its follower a source of illumination that may be applied from different angle. |
Interior documentation in Saint Nicholas at Old Town Square in Prague.Deployment of the system of two cooperating MAVs in Saint Nicholas Church at Old Town Square in Prague. The light provided from the angle of 45 degrees enables to introduce deepness into the gained images. This allows to present details of statues and other 3D objects in the churches. |
Predictive control and stabilization of a formation of MAVs filming in dark conditions.The leader is equipped by a camera and followers by light sources. |
Object detection for 'Treasure hunt' challenge of the MBZIRC 2017 contest.The video demonstrates the object detection and tracking for a quadrotor UAV. The UAV's task was to pick up colored objects and deliver them to a specific drop-off zone. For details, see http://mrs.felk.cvut.cz/projects/mbzirc. During the first 2 minutes, the UAV flies over the field at a high altitude and creates a map of the objects in the contest area. Then it creates a plan for the object pickup, descends to a lower altitude and starts picking the objects. There are other two UAV's operating in the arena and this UAV has to coordinate with them in order to perform the task efficiently. Note that the method also estimates the velocity of the objects, so that the UAV can pick up the objects from a moving robot (see 4:35). Note the absence of false positives and excellent detection rate despite the jelly effect caused by the drones rotors. |
Fully decentralized swarming behavior in an artificial forest.An experiment of three self-localised MAVs flying through an artificial forest. MAVs control is fully decentralised based on information about local neighbours of the robots. |
Radiation source localization by a formation of MAVs.Localization of a radiation source by a formation of MAVs based on detecting intensity of the simulated source of radiation. |
Realistic simulation of a compact self-stabilized MAV group forming a distributed sensory array.A demonstration of the main ideas in the proposed project. A realistic simulation of a compact self-stabilized MAV group forming a distributed sensory array to illustrate the proposed concept of flying adaptive “antennas”. The flexible sensory array adapts its shape to move through an environment with obstacles. The realistic simulations (including MAV dynamics and interaction with environment) were prepared in Gazebo using the multi-MAV control system successfully deployed by our team in the MBZIRC competition. |
Cooperative carriage of large objects by a pair of MAVs.The robots are relatively localized by onboard vision mechanism. |
Autonomous drive at night using BearNav navigation system.The video demonstrates the capability of the BearNav visual navigation system (https://github.com/gestom/stroll_bearnav) to guide a mobile robot in particularly difficult lighting conditions. |
Virtual migrating leader real experiment.A compact formation of micro aerial vehicles flying through an outdoor environment with obstacles using a migrating virtual leader approach. |
UAV recognition using Convolutional neural network.Convolutional neural network (CNN) technique, which is in development by CTU for localization of MAVs in adverse light conditions. A concept of smart lightening by an autonomous formation of MAVs carrying a camera and independent light sources in historical objects was used as a motivation of this experiment. |
Disturbing UAV by adding additional weight.The experiment shows capability of the UAV to reject disturbances induced by adding an additional weight of-center of its center of gravity. |
List of submitted publications related to the MBZIRC 2017 proposal. Please, consider the attached pdf files of manuscripts as confidential, since they are under review: |
|
|
|
