While deep reinforcement learning (RL) methods have achieved unprecedented successes in a range of challenging problems, their applicability has been mainly limited to simulation or game domains due to the high sample complexity of the trial-and-error learning process. However, real-world robotic applications often need a data-efficient learning process with safety-critical constraints. In this paper, we consider the challenging problem of learning unmanned aerial vehicle (UAV) control for tracking a moving target. To acquire a strategy that combines perception and control, we represent the policy by a convolutional neural network. We develop a hierarchical approach that combines a model-free policy gradient method with a conventional feedback proportional-integral-derivative (PID) controller to enable stable learning without catastrophic failure. The neural network is trained by a combination of supervised learning from raw images and reinforcement learning from games of self-play. We show that the proposed approach can learn a target following policy in a simulator efficiently and the learned behavior can be successfully transferred to the DJI quadrotor platform for real-world UAV control.

Obtaining reliable state estimates at high altitude but GPS-denied environments, such as between high-rise buildings or in the middle of deep canyons, is known to be challenging, due to the lack of direct distance measurements. Monocular visual-inertial systems provide a possible way to recover the metric distance through proper integration of visual and inertial measurements. However, the nonlinear optimization problem for state estimation suffers from poor numerical conditioning or even degeneration, due to difficulties in obtaining observations of visual features with sufficient parallax, and the excessive period of inertial measurement integration. In this paper, we propose a spline-based high altitude estimator initialization method for monocular visual-inertial navigation system (VINS) with special attention to the numerical issues. Our formulation takes only inertial measurements that contain sufficient excitation, and drops uninformative measurements such as those obtained during hovering. In addition, our method explicitly reduces the number of parameters to be estimated in order to achieve earlier convergence. Based on the initialization results, a complete closed-loop system is constructed for high altitude navigation. Extensive experiments are conducted to validate our approach.

We present an online method for generating collision-free trajectories for autonomous quadrotor flight through cluttered environments. We consider the real-world scenario that the quadrotor aerial robot is equipped with limited sensing and operates in initially unknown environments. During flight, an octree-based environment representation is incrementally built using onboard sensors. Utilizing efficient operations in the octree data structure, we are able to generate free-space flight corridors consisting of large overlapping 3-D grids in an online fashion. A novel optimization-based method then generates smooth trajectories that both are bounded entirely within the safe flight corridor and satisfy higher order dynamical constraints. Our method computes valid trajectories within fractions of a second on a moderately fast computer, thus permitting online re-generation of trajectories for reaction to new obstacles. We build a complete quadrotor testbed with onboard sensing, state estimation, mapping, and control, and integrate the proposed method to show online navigation through complex unknown environments.

In this work, we address the problem of aggressive flight of a quadrotor aerial vehicle using cameras and IMUs as the only sensing modalities. We present a fully integrated quadrotor system and demonstrate through online experiment the capability of autonomous flight with linear velocities up to 4.2 m/s, linear accelerations up to 9.6 m/s2, and angular velocities up to 245.1 degree/s. Central to our approach is a dense visual-inertial state estimator for reliable tracking of aggressive motions. An uncertainty-aware direct dense visual tracking module provides camera pose tracking that takes inverse depth uncertainty into account and is resistant to motion blur. Measurements from IMU pre-integration and multi-constrained dense visual tracking are fused probabilistically using an optimization-based sensor fusion framework. Extensive statistical analysis and comparison are presented to verify the performance of the proposed approach. We also release our code as open-source ROS packages.

We address the challenging problem of tracking a moving target in cluttered environments using a quadrotor. Our online trajectory planning method generates smooth, dynamically feasible, and collision-free polynomial trajectories that follow a visually-tracked moving target. As visual observations of the target are obtained, the target trajectory can be estimated and used to predict the target motion for a short time horizon. We propose a formulation to embed both limited horizon tracking error and quadrotor control costs in the cost function for a quadratic programming (QP), while encoding both collision avoidance and dynamical feasibility as linear inequality constraints for the QP. Our method generates tracking trajectories in the order of milliseconds and is therefore suitable for online target tracking with a limited sensing range. We implement our approach on-board a quadrotor testbed equipped with cameras, a laser range finder, an IMU, and onboard computing. Statistical analysis, simulation, and real-world experiments are conducted to demonstrate the effectiveness of our approach.

We address the important problem of achieving robust and easy-to-deploy visual state estimation for micro aerial vehicles (MAVs) operating in complex environments. We use a sensor suite consisting of multiple cameras and an IMU to maximize perceptual awareness of the surroundings and provide sufficient redundancy against sensor failures. Our approach starts with an online initialization procedure that simultaneously estimates the transformation between each camera and the IMU, as well as the initial velocity and attitude of the platform, without any prior knowledge about the mechanical configuration of the sensor suite. Based on the initial calibrations, a tightly-coupled, optimization-based, generalized multi-camera-inertial fusion method runs onboard the MAV with online camera-IMU calibration refinement and identification of sensor failures. Our approach dynamically configures the system into monocular, stereo, or other multi-camera visual-inertial settings, with their respective perceptual advantages, based on the availability of visual measurements. We show that even under random camera failures, our method can be used for feedback control of the MAVs. We highlight our approach in challenging indoor-outdoor navigation tasks with large variations in vehicle height and speed, scene depth, and illumination.