How computer scientists can advance robotics
Fuelled by a recent surge in artificial intelligence technologies, robotics is on the verge of taking centre stage in our society. Nevertheless, before robots can assist us with everyday chores at work and at home, they will need the skills to perform an increasingly diverse array of tasks. Professor Stelian Coros, leader of the Computational Robotics Lab within the Department of Computer Science, is developing the mathematical foundations for new algorithms that address this fundamental challenge.
Professor Coros wants to advance the field of robotics by making its problems more accessible to computer science students. “Because of my background in computer science, I don’t approach robotics problems in a traditional way. For example, when I think of creating new types of robots, I imagine an algorithmic approach that leverages computational resources rather than typical, largely manual design methodologies,” the assistant professor explains. This marriage of computer science thinking and robotics could significantly increase the pace of progress.
Smarter and safer
In their quest to creating next-gen robots with rich repertoires of behaviours, Professor Coros and his collaborators study the principles of dexterous manipulation and legged locomotion. To this end, they recently developed computational motor control models that allow an external pageintelligent multi-limbed robotcall_made to seamlessly choose whether to use its limbs as legs or as arms. This unique feature allows the robot to transition between nimble locomotion and configurations that allow it to grasp and carry objects. In addition to becoming increasingly versatile, robots that interact with people or handle fragile objects must have a soft touch. Aiming for robots that are inherently safe, Professor Coros and his collaborators are developing Downloadsoft anthropomorphic manipulators (MOV, 28.3 MB)vertical_align_bottom that can dexterously grasp and manipulate a variety of objects.
Robots tailored to specific tasks
Imagine a scenario where scientists are studying a remote ecosystem. As part of their investigations, they would like to deploy a robot for a new inspection task. Depending on the type of task, considerably different solutions may be necessary. For example, if soil samples need to be collected, the robot should be designed to carry a potentially significant payload; monitoring air quality over longer periods of time would require the robot to be lightweight and energy efficient; the need to traverse challenging terrains, access hard-to-reach places or clear debris, on the other hand, would impose constraints on the robot’s size, its ability to move nimbly, or the workspace and dexterity of its end-effectors. To undertake both large and small tasks as they arise, the scientists would greatly benefit from the ability to create different types of robots directly on site. Thanks to the research being conducted by Professor Coros, this vision could soon become a reality.
The challenges of creating robots
Creating robots is notoriously difficult. Based on the tasks a robot is intended for, designers must consider an overwhelming number of design choices, including the size, shape and material composition of each body part, the selection and layout of sensors and actuators, and the choice of control algorithms. Standard engineering practices approach this daunting design challenge through extensive manual experimentation – a workflow that is slow, expensive and prone to errors. As a consequence, it is not currently possible to create task-specific robots on demand, and today’s design methodologies will not scale as demand grows for an increasingly diverse ecosystem of robots.
Intelligent software to produce customised hardware
Professor Coros and his team are taking an engineering meets AI approach to creating bespoke robots: they are developing the algorithmic foundation for software that can design customised hardware. With such software, robots can be easily customised for different tasks, enabling developers to focus on creative innovation rather than tedious fine-tuning. This new paradigm for designing robots is exemplified by the Skaterbots project. Through an intuitive graphical user interface, designers can create unique robots by combining different types of actuators, 3D printable connectors, wheels and feet in a mix-and-match manner. The bespoke robots designed in this way then automatically learn how to walk, roll, glide or skate without requiring any programming from the designer. “To enable the vision I have regarding the future of robotics, advances in hardware and software systems must complement each other synergistically,” says Professor Coros.