Can Micro Aerial Vehicles (MAVs) Fly in Woods Like Birds?
The City College of New York, CCNY Robotics Lab
Abstract:
In recent years, many research teams have achieved autonomous navigation of Micro Aerial Vehicles (MAVs) in indoor environments. However, it remains an unsolved and extremely difficult problem for MAVs to operate in cluttered and obstacle-dense environments such as caves or woods. The difficulties stem from a number of factors, including limited payload, insufficient on-board sensing and information processing capabilities, and unstable flight control in response to disturbances (e.g. wind gusts) and payload variations.
Leveraging past achievements in 3D SLAM of quadrotor UAV in rectilinear indoor environments, the robotics team at The City College of New York propose to tackle these technical challenges to push the limit in UAV research. The team will continue the efforts to design special light-weight perceptual sensors (i.e., single camera omni-vision system, 3D laser scanner, Kinect-based RGB-D sensor) for MAV navigation. By fusing the information from these perceptual sensors with the existing IMU, GPS and altimeter sensors on-board the AscTec UAVs, the team will develop sophisticated control and navigation algorithms to empower the UAVs with the similar exploration capability of flying birds to achieve autonomous navigation in wooded environments.
The team will share the research results and contribute open source software packages to ROS.org community as they did before (http://www.ros.org/wiki/ccny-ros-pkg).
AscTec UAVs in the Wild
University of Nebraska-Lincoln, Nebraska Intelligent MoBile Unmanned Systems (NIMBUS) Laboratory
Abstract:
We plan to develop the hardware, software, and algorithms to enable an AscTec UAV to catch and avoid balls that are thrown at it. Consider a group of kids all throwing balls at the UAV from all angles. The UAV would avoid being hit by red balls, while catching green balls. This is the first step in our long-term goal of creating systems and algorithms that will enable UAVs to reach their full potential in dynamic, real-world environments.
Sensing and catching balls requires real-time vision processing, obstacle avoidance, and dynamic motion planning. The AscTec UAVs have the power and agility to achieve these goals. Their agility has been illustrated time and again in motion capture rooms. Currently, without this precise localization information, researchers limit autonomously operating UAVs to slow and deliberate actions to avoid running into obstacles.
We aim to bridge this gap to enable high-speed, dynamic actions on UAVs in the real-world. The real-world is messy; there is no motion capture system, GPS and radio may be intermittent, wind is a factor, and there are many dynamic obstacles. Catching balls will show that the UAVs have the processing capabilities and agility to autonomously interact with real-time physical events in the real-world. We plan to develop a lighter weight high speed and high resolution camera processing system to support the needed vision processing, which will leave sufficient thrust to allow high-speed dynamic actions. We will demonstrate the system at a variety of public events, including at a nationally televised UNL sporting events.
