Speakers from Academia

Talk Title: SMT-based Planning - Some recent developments

Planning is a highly challenging central module in robotics systems. For high-level planning, besides traditional approaches also SMT solving has been shown to be a good alternative solution. In this talk, we report on some recent developments and ideas in this area.

Talk Title: Knowledge representation & reasoning in CRAM-a cognitive architecture for robot agents accomplishing underdetermined manipulation tasks

Robotic agents that can accomplish manipulation tasks with the competence of humans have been one of the grand research challenges for artificial intelligence (AI) and robotics research for more than 50 years. However, while the fields made huge progress over the years, this ultimate goal is still out of reach. I believe that this is the case because the knowledge representation and reasoning methods that have been proposed in AI so far are necessary but too abstract. In this talk I propose to address this problem by endowing robots with the capability to internally emulate and simulate their perception-action loops based on realistic images and faithful physics simulations, which are made machine-understandable by casting them as virtual symbolic knowledge bases. These capabilities allow robots to generate huge collections of machine-understandable manipulation experiences, which robotic agents can generalize into commonsense and intuitive physics knowledge applicable to open varieties of manipulation tasks. The combination of learning, representation, and reasoning will equip robots with an understanding of the relation between their motions and the physical effects they cause at an unprecedented level of realism, depth, and breadth, and enable them to master human-scale manipulation tasks. This breakthrough will be achievable by combining leading-edge simulation and visual rendering technologies with mechanisms to semantically interpret and introspect internal simulation data structures and processes. With this talk we will give an outlook of applying CRAM to underwater missions.

Talk Title: Autonomy in marine robotics - AI or automation?

The talk will first give a brief overview on current robotic systems in the maritime context, from floaters and gliders over remotely operated vehicles (ROVs), hybrid ROVs, to autonomous underwater vehicles (AUVs) and subsea crawlers with different grades of autonomy. We will then take a deeper look into the software architectures and autonomous capabilities of such systems and discuss autonomous mission planning, task decomposition and execution, concepts like behaviour trees, approaches to sensor fusion and environment representation and how systems can reason about their own state and marine environment and act accordingly. The talk will also showcase two real-world examples of successful application of machine learning methods to marine robots in order to identify the motion model and to learn magnetic field distortions for navigation purposes. During the whole talk, we try to address the question, how far we have gotten in the field of marine robotics towards truly autonomous systems and behaviours.

Talk Title: The Long Way to Persistent Autonomy in Underwater Robotics

The future vision for ocean observatories foresees a multitude of permanently deployed autonomous vehicles, communicating with each other, and having intelligent decision-making capabilities in navigation, energy management, and error handling. There is however a stark contrast with respect to current capabilities: commercial vehicles are very good in blindly executing pre-defined paths, they easily get stuck when conditions change and limited on-board decision making severely limits the types and length of missions. What are the main challenges to bridge our future vision with the current reality? This talk will present the results of various projects in the last few years, aiming at increasing vehicles' capabilities. The role of a dynamically updated probabilistic knowledge-base opens up the door to the use of AI techniques like KR and reasoning. Classical AI Planning can be applied to robotics missions, instead of using efficient but not very adaptable state machines. Recent advances in Deep Learning can provide important improvements in sensor processing and object recognition. Adding on-board sensor-processing, decision making, fault management, underwater vehicles seem to significantly move forward in the field of autonomy. However, how promising research results, which are still validated during relatively controlled conditions, can be generalised to produce a real impact in the field?

Speakers from Industry

Talk Title: Serving AI for underwater applications

EIVA software team focuses on developing sensor and platform-agnostic solutions for underwater applications to accommodate our customers' needs. Research in AI offers tools and methods that can provide invaluable automation possibilities for underwater engineering and survey, and thus we constantly evaluate and use new AI developments. In this talk, an AI-based solution for automating underwater pipeline inspection will be presented, which is a mature but constantly improving product by EIVA. It was initially deployed to our customers three years ago. The employed development methodology is now considered to be an AI industry standard. It is a data-hungry and computationally intensive approach that requires periodic adjustments but yields consistent high-quality results under a variety of conditions. We will look into the advantages and limitations of this method.

Talk Title: A roadmap towards integrating results from machine learning in online automated decision making for autonomous underwater robots

Besides a well-tuned network setup, the quality of processing results from machine learning algorithms depends to a large extent on the quality of the training dataset. If autonomous robots face situations where there is a lack of training data, there is no human to take over control like for assisted driving cars. This is especially true for underwater robots where a connection to a supervisor on land is limited by the bandwidth of underwater acoustics. Thus, misclassified results can lead to fatal decisions which have to be avoided at all costs especially for underwater robots as these often operate in critical environments. This talk approaches this topic and proposes some ways to start integrating results from machine learning algorithms into the autonomy of underwater robots.

Talk Title: Towards fully autonomous surveys with teams of autonomous surface and underwater vehicles

At Maritime Robotics, we are developing autonomous system solutions for the maritime domain. Primarily, our focus has been on developing autonomous surface vehicles (ASVs), that are used with a suite of sensors for seafloor mapping and environmental monitoring. The recent pandemic has highlighted some of the benefits of using marine robots; requiring fewer crew for remote or autonomous operations. For operations with autonomous, and remotely operated, underwater vehicles, we see that currently crewed surface vessels are typically needed for deployment and recovery, as well as for overseeing operations. In this talk, I will briefly show some of the research projects that we are involved in, which aim at enabling ASVs to work in support of underwater vehicles, for data sharing and improved underwater localization, as well as future autonomous launch and recovery. Apart from putting demands on the ASV specific to the application, it also requires that we push the level of autonomy on the ASV itself.

Talk Title: Event-based vision and neuromorphic computing for autonomous robots

Event-based cameras are low-latency, low-power, low data rate, and high dynamic range visual sensors that harness the properties of biologically-inspired sensing and computing. These characteristics can be exploited in challenging environments, reducing motion blur and coping with conditions of varying contrast in the field of view. When combined with neuromorphic computing hardware that features efficient computing of event-based (spiking) neuronal networks and scalable on-chip plasticity, event-based vision and computing could lead to a new family of computing systems that will enable low-power and low-data AI for robotic agents working in challenging environments, including underwater robotics.

Talk Title: Underwater Perception with Laser Scanners

Laser scanners are readily available for terrestrial applications and provide the basis for many autonomous systems. This is not easily transferred to underwater applications where optical sensors face multiple challenges. Given the right conditions laser scanners can provide important foundations for the perception pipelines for AI in underwater applications.