Drones are expected to be used extensively for delivery tasks in the future. In the absence of obstacles, satellite based navigation from departure to the geo-located destination is a simple task. When obstacles are known to be in the path, pilots must build a flight plan to avoid them. However, when they are unknown, there are too many or they are in places that are not fixed positions, then to build a safe flight plan becomes very challenging. Moreover, in a weak satellite signal environment, such as indoors, under trees canopy or in urban canyons, the current drone navigation systems may fail. Artificial intelligence, a research area with increasing activity, can be used to overcome such challenges. Initially focused on robots and now mostly applied to ground vehicles, artificial intelligence begins to be used also to train drones. Reinforcement learning is the branch of artificial intelligence able to train machines. The application of reinforcement learning to drones will provide them with more intelligence, eventually converting drones in fully-autonomous machines. In this work, reinforcement learning is studied for drone delivery. As sensors, the drone only has a stereo-vision front camera, from which depth information is obtained. The drone is trained to fly to a destination in a neighborhood environment that has plenty of obstacles such as trees, cables, cars and houses. The flying area is also delimited by a geo-fence; this is a virtual (non-visible) fence that prevents the drone from entering or leaving a defined area. The drone has to avoid visible obstacles and has to reach a goal. Results show that, in comparison with the previous results, the new algorithms have better results, not only with a better reward, but also with a reduction of its variance. The second contribution is the checkpoints. They consist of saving a trained model every time a better reward is achieved. Results show how checkpoints improve the test results.

Unmanned aerial vehicles (UAV) specifically drones have been used for surveillance, shipping and delivery, wildlife monitoring, disaster management etc. The increase on the number of drones in the airspace worldwide will lead necessarily to full autonomous drones. Given the expected huge number of drones, if they were operated by human pilots, the possibility to collide with each other could be too high.In this paper, deep reinforcement learning (DRL) architecture is proposed to make drones behave autonomously inside a suburb neighborhood environment. The environment in the simulator has plenty of obstacles such as trees, cables, parked cars and houses. In addition, there are also another drones, acting as moving obstacles, inside the environment while the other drone has a goal to achieve. In this way the drone can be trained to detect stationary and moving obstacles inside the neighborhood and so the drones can be used safely in a public area in the future. The drone has a front camera and it can capture continuously depth images. Every depth image, with a size of 144x256 pixels, is part of the state used in DRL architecture. Also, another part of the state is the distance to the geo-fence, a virtual barrier on the environment, which is added as a scalar value. The agent will be rewarded negatively when it tries to overpass the geo-fence limits. In addition, angle to goal and elevation angle between the goal and the drone will be used as information to be added to the state. It is considered that these scalar values will improve the DRL performance and also the reward obtained. The drone is trained using Q-Network and its convergence and final reward are evaluated. The states containing image and several scalars are processed by neural network that joints the two state parts into a unique flow. This neural network is named as Joint Neural Network (JNN) [1]. The training and test results show that the agent can successfully avoid any obstacles in the environment. In training, there exist some episodes crashed at the beginning of the training session because the random drones moves randomly in the environment and thus they can hit the learner drone during training. The test results are very promising and the learner drone reaches the destination with a success rate %100 in first two tests and with a success rate %98 in the last test.