At the Big Data & AI Leaders Summit Boston 2018, Ryan Sinnet from Miso Robotics delivered his presentation on the modernisation of food preparation using AI.

Miso Robotics is a tech startup that began in 2016 with a vision to transform the commercial kitchen through technology. They have since created Flippy, world’s first autonomous robotic kitchen assistant. Flippy is an autonomous kitchen assistant that the company hopes will take over repetitive tasks, allowing staff in restaurants to focus on hospitality, and to experience growth and savings.

Flippy was widely covered in the media and they began serving the public this year in Caliburger, running the lunch service. Miso Robotics hopes to continue to expand, develop more AI robotic systems and allow Flippy to move beyond burgers.

Ryan discussed Miso Robotic’s approach to development: a triangle that combines mechanical design, artificial intelligence and formal methods.

He explains that while mechanical design is an old school approach he feels that it can still be highly effective, as some problems in software are extremely difficult to code around. Ryan offered the example of how, for Flippy, they chose to give the robot specialised grippers, rather than a fully mechanical hand with five fingers, for simplicity’s sake.

Miso Robotics wants to develop robots that are automatically adaptable in any situation, which is why Flippy uses artificial intelligence as it allows it to learn independently. However, Ryan notes that whilst using AI allows adaptability, it’s not always the best solution and formal methods and control theories are still needed as it provides provable guarantees on performance.

Ryan stresses the importance of balance: when solving a problem, one must choose a solution that blends all three of these approaches without drifting too much in one corner of the triangle. Additionally, it is necessary to achieve rapid commercialisation but without sacrificing scalability.

Looking towards the future, Ryan predicts that “the next generation of autonomous machines will use artificial intelligence and computer vision with modern control theory to solve complex tasks in varied environments”. Adaptability is a key issue that those in AI must face - previously, robots were designed with specialised tools to solve specific, repetitive tasks. While this system worked in factories where the environment is controlled, if there are any changes or if the robot is deployed in a different environment, it would struggle.

Ryan posed three problems that are obstacles to building artificially intelligent, autonomous machines.

  1. The perception problem: how can a machine using a variety of sensors meaningfully understand the world around it? Currently the development of AI has been a big advancement, as these systems can be taught by providing them with information so they can interpret and learn.

  2. The manipulation problem: how can a machine interact with its environment to achieve a desired outcome? Ryan notes that by combining AI with modern control theory, new capabilities are brought to robots.

  3. The decision problem: how can a machine act to navigate a variety of scheduling constraints while interacting with external agents? Autonomous machines need to be able to perform a variety of actions and know when to perform each action, and they are able to make highly optimised decisions if they has access to a lot of information and compute power.

In order to move towards a future of developing fully autonomous and artificially intelligent robots that can adapt to their environment, problems will need to be faced by combining different approaches of mechanical design, AI and formal methods.

In the meantime, Flippy will be at the Dodgers Stadium, working the frying stations to serve fried chicken and tater tots to customers.

It’s easy to get onboard and start benefitting instantly. Either enrol at a Summit or sign up for an annual membership.