Neuromorphic hardware is capable of processing sensory data from Dynamic Vision Sensors (DVS) and Event‐based Cameras at the scale of micro‐seconds (or at the rate of MHz). This capability has been demonstrated recently through experimentation on benchmark experimental platforms[1‐3]. Thisis great news for the robotics and AI research community focused on autonomous agile aerial robotics, where the sensory data needs to be processed at very high speeds. Moreover, neuromorphic computing is expected to show tremendous gains over conventional computing, in terms of latency and energy savings, in several areas involving learning, optimization and control [4,5].

In this Master Thesis being proposed, we wish to explore the potential of neuromorphic computing for accomplishing agile maneuvers using aerial robots. Along the way, we also wish to evaluate the latencies introduced at the interfaces between neuromorphic computing in the cloud and platforms such as ROS and Gazebo. Currently,severalsoftware packages and intermediary hardware are being used asinterfaces between conventional computing and neuromorphic computing and research is underway that aims to develop smoother interfaces. The evaluation would help in getting an estimate of the actual speeds at which the feedback loops can be closed, with all the necessary interfaces in place.

Some intermediate tasks leading up to the final goals would be the following:

1. Reproducing recent results which use neuromorphic computing for aerial robotics, to gauge the state‐of‐the‐art, while also gaining the basic working knowledge of neuromorphic computing.

2. Interfacing neuromorphic sensorssuch as Prophesee and DVS with Intel neuromorphic computing in the cloud. Becoming conversant with ROS and Gazebo packages that have been developed recently for interfacing with Intel Loihi.

3. Evaluation of latencies introduced at the interfaces between ROS and Gazebo platforms and neuromorphic computing. The interfaces are expected to induce significant delays and we look to design interfaces which are smoother and more direct.  

4. Implementing model‐based and machine learning based control algorithms in the neuromorphic computing cloud, to realize agile aerial maneuvers. Experimental demonstration on the aerial platforms that are part of the Robotics and AI group at LTU.