The application of deep learning in robotics leads to very specific problems and research questions that are typically not addressed by the computer vision and machine learning communities. In this paper we discuss a number of robotics-specific learning, reasoning, and embodiment challenges for deep learning. We explain the need for better evaluation metrics, highlight the importance and unique challenges for deep robotic learning in simulation, and explore the spectrum between purely data-driven and model-driven approaches. We hope this paper provides a motivating overview of important research directions to overcome the current limitations, and helps to fulfill the promising potentials of deep learning in robotics.

https://journals.sagepub.com/doi/pdf/10.1177/0278364918770733

The limits and potentials of deep learning for robotics

A compilation of UAV applications for precision agriculture

Optimal State Estimation

where xt+1 is the position and orientation of the robot (with respect to a reference frame) at time t + 1, with (ξ, η) giving the x and y coordinates and θ the angle (with respect to the x-axis) that the robot is facing. The robot has two drive wheels, of radius r on an axle of length d. During time period t the right wheel is believed to rotate at a velocity of φRt and the left at a velocity of φLt. In this example, these velocities are fixed with φRt = 0.4 and φLt = 0.1. The state update function, F , calculates where the robot should be at each time point, given its previous position. However, in reality, there is some random fluctuation in the velocity of the wheels, for example, due to slippage. Therefore the actual position of the robot and the position given by equation F will differ. The magnitude of these random fluctuations is included in the model through Σx.

Unscented Kalman Filterの計測への応用に関する研究

In this letter, we present a method to control a quadrotor with a neural network trained using reinforcement learning techniques. With reinforcement learning, a common network can be trained to directly map state to actuator command making any predefined control structure obsolete for training. Moreover, we present a new learning algorithm that differs from the existing ones in certain aspects. Our algorithm is conservative but stable for complicated tasks. We found that it is more applicable to controlling a quadrotor than existing algorithms. We demonstrate the performance of the trained policy both in simulation and with a real quadrotor. Experiments show that our policy network can react to step response relatively accurately. With the same policy, we also demonstrate that we can stabilize the quadrotor in the air even under very harsh initialization (manually throwing it upside-down in the air with an initial velocity of 5 m/s). Computation time of evaluating the policy is only 7 $\mu$ s per time step, which is two orders of magnitude less than common trajectory optimization algorithms with an approximated model.

Control of a Quadrotor With Reinforcement Learning

This paper presents a novel path planning algorithm for the autonomous exploration of unknown space using aerial robotic platforms The proposed planner employs a receding horizon “next-best-view” scheme: In an online computed random tree it finds the best branch, the quality of which is determined by the amount of unmapped space that can be explored Only the first edge of this branch is executed at every planning step, while repetition of this procedure leads to complete exploration results The proposed planner is capable of running online, onboard a robot with limited resources Its high performance is evaluated in detailed simulation studies as well as in a challenging real world experiment using a rotorcraft micro aerial vehicle Analysis on the computational complexity of the algorithm is provided and its good scaling properties enable the handling of large scale and complex problem setups

Receding horizon “next-best-view” planner for 3D exploration

This paper presents a framework for real-time, full-state feedback, unconstrained, nonlinear model predictive control that combines trajectory optimization and tracking control in a single, unified approach. The proposed method uses an iterative optimal control algorithm, namely Sequential Linear Quadratic (SLQ), in a Model Predictive Control (MPC) setting to solve the underlying nonlinear control problem and simultaneously derive the optimal feedforward and feedback terms. Our customized solver can generate trajectories of multiple seconds within only a few milliseconds. The performance of the approach is validated on two different hardware platforms, an AscTec Firefly hexacopter and the ball balancing robot Rezero. In contrast to similar approaches, we perform experiments that require leveraging the full system dynamics.

Fast nonlinear Model Predictive Control for unified trajectory optimization and tracking

In this chapter, strategies for Model Predictive Control (MPC) design and implementation for Unmaned Aerial Vehicles (UAVs) are discussed. This chapter is divided into two main sections. In the first section, modelling, controller design and implementation of MPC for multi-rotor systems is presented. In the second section, we show modelling and controller design techniques for fixed-wing UAVs. System identification techniques are used to derive an estimate of the system model, while state of the art solvers are employed to solve the optimization problem online. By the end of this chapter, the reader should be able to implement an MPC to achieve trajectory tracking for both multi-rotor systems and fixed-wing UAVs.

Model Predictive Control for Trajectory Tracking of Unmanned Aerial Vehicles Using Robot Operating System

Within this paper a new path planning algorithm for autonomous robotic exploration and inspection is presented. The proposed method plans online in a receding horizon fashion by sampling possible future configurations in a geometric random tree. The choice of the objective function enables the planning for either the exploration of unknown volume or inspection of a given surface manifold in both known and unknown volume. Application to rotorcraft Micro Aerial Vehicles is presented, although planning for other types of robotic platforms is possible, even in the absence of a boundary value solver and subject to nonholonomic constraints. Furthermore, the method allows the integration of a wide variety of sensor models. The presented analysis of computational complexity and thorough simulations-based evaluation indicate good scaling properties with respect to the scenario complexity. Feasibility and practical applicability are demonstrated in real-life experimental test cases with full on-board computation.

Receding horizon path planning for 3D exploration and surface inspection

Extending the maneuverability of multirotors promises to yield a considerable increase in their scope of applications, such as carrying out more challenging inspection tasks and generating complex, uninterrupted movements of an attached camera. In this article, we address the promise of multirotor maneuverability by presenting Voliro, a novel aerial platform that combines the advantages of existing multirotor systems with the agility of vehicles having omniorientational controllability. In other words, Voliro can fly in any direction while maintaining an arbitrary orientation.

Mina Kamel

Papers

Receding horizon “next-best-view” planner for 3D exploration

Fast nonlinear Model Predictive Control for unified trajectory optimization and tracking

Model Predictive Control for Trajectory Tracking of Unmanned Aerial Vehicles Using Robot Operating System

Receding horizon path planning for 3D exploration and surface inspection

The Voliro Omniorientational Hexacopter: An Agile and Maneuverable Tiltable-Rotor Aerial Vehicle