A robot “chef” was taught to taste food at various stages of the chewing process to determine if it was seasoned enough. Researchers from the University of Cambridge, together with home appliance manufacturer Beko, trained the robot chef to judge the saltiness of a food at different stages of the chewing process, simulating a comparable process in humans.
Their findings could help improve automatic or semi-automatic food preparation by helping robots learn what is delicious and what is bad, thereby improving their cooking skills.
As we chew our food, we detect a difference in texture and flavor. For example, biting into a fresh tomato in the summer releases its juices and as chewing, saliva and digestive enzymes are released, our impression of the tomato's flavor changes.
The robot chef, who has already been trained to make omelets based on human feedback, tasted nine different kinds of omelets and tomatoes at three different stages of the chewing process and developed "taste maps" of various dishes.
Compared with existing electronic tasting technologies that only evaluate a single homogenized sample, the researchers discovered that this "taste as you go" method significantly increases the robot's capacity to quickly and precisely predict the saltiness of food. Frontiers in Robotics and AI published the findings.
Taste perception is a multi-step process that has evolved in humans over millions of years: The appearance, smell, texture, and temperature of food affect how people experience taste; saliva formed during chewing helps to transport chemical compounds in food to taste receptors located largely on the tongue; and signals from taste buds are sent to the brain. When our brain learns its flavor, we decide whether we like food or not.
Taste is also highly personal: some people prefer spicy dishes, while others prefer sweets. Whether amateur or expert, a good cook uses his sense of taste to balance the flavors in a dish and produce a well-rounded end result.
"Most home cooks are familiar with the concept of tasting by checking a food throughout the cooking process to see if the flavor balance is perfect," said Grzegorz Sochacki, the paper's first author from the Cambridge Engineering Department. If they are to be used in certain food preparation areas, it is critical that the robots be able to 'taste' what they are cooking.”
Also from the Engineering Department, Dr. “When we taste it, the chewing process also sends constant inputs to our brains,” said Arsen Abdulali. “Since current electronic testing methods take only a single snapshot of a homogenized sample, we aimed to mimic a more realistic chewing and tasting process in a robotic system, resulting in a tastier end product.”
The researchers are part of Cambridge's Laboratory of Bio-Inspired Robotics, led by Professor Fumiya Iida of the Department of Engineering, which focuses on teaching robots to solve so-called last-meter problems that humans find simple but difficult for robots to solve. One of those tasks is cooking: Earlier testing with robots “chef” yielded a good omelet using human taster feedback.
“We wanted something cheap, compact and quick to add to our robot so it could taste,” Sochacki said.
The researchers attached a conductivity probe that works as a salinity sensor to a robot arm to mimic the human chewing and tasting process in robot chefs. They made the scrambled eggs and tomatoes with a different amount of tomatoes and salt on each plate.
The robot “tasted” the plates with the probe in a grid-like pattern and produced a reading in just a few seconds.
The scientists stirred the egg mixture to simulate the textural change caused by chewing, and had the robot test the plate one more time. Taste maps of each meal were created using different readings at different "chew" times.
Their findings revealed that the robots were much better at assessing salinity than other electronic tasting methods, which often take time and provide only a single reading.
While the researchers' method is still a proof-of-concept, they believe that robots may one day be able to mimic the processes of chewing and tasting humans to produce food that humans will enjoy and that can be tailored to personal tastes.
“When a robot learns to cook, like any normal cook, it needs feedback on how well it is doing,” Abdulali said. “We want robots to grasp the concept of taste so they can be great cooks.”
The robot can see the differences in the food eaten in our experiment, which increases its capacity for taste.”
Senior Scientist from Beko, Dr. “Beko's mission is to bring safe and easy-to-use robots into the home,” said Muhammad W. Chughtai. “We expect robot chefs to play an important role in busy homes and assisted care facilities in the future.” This is an important step forward in robotic cooking as chewing will allow robot chefs to tailor flavor to different dishes and users using machine and deep learning techniques.”
The researchers hope to improve the robot chef in the future so that he can taste a variety of foods and enhance his sensory abilities, for example, to detect sweet or fatty foods.