Bicycles are among the biggest challenges of self-driving robotic cars
Q. This musical maestro of the insect world uses its various body parts to create unique sounds. Can you name this tunemaker? A. It’s the tiny masked birch caterpillar (Drepana arcuata) that shakes its body, drums and scrapes its mouthparts, and drags specialized anal “oars” against the leaf surface to create bizarre signals, says biologist Jayne Yack, as reported in “New Scientist” magazine. It may be that these signals serve to invite other caterpillars to food-rich areas for group shelter-building (Behavioral Ecology and Sociobiology). After more than 30 years of studying insect sounds, Yack declares, “I’ve never seen one insect species produce such a diversity of signal types.” Q. The technology behind self-driving “robotic” vehicles is getting better and better at detecting other cars, trucks and vans, and now even pedestrians, squirrels and birds. But it still faces a daunting challenge from the lightest and quietest vehicles. Got a guess? A. Add the swerviest vehicles and you’ll likely know that bicycles are the answer. Enclosed vehicles are big and easy to see and their pathways are easy to predict, says Peter Fairley in “IEEE Spectrum” magazine. Bicycles, on the other hand, are small, fast and heterogeneous. As visual computing expert Nuno Vasconcelos explains, “A car is basically a big block of stuff. A bicycle has much less mass… There are more shapes and colors, and people hang stuff on them.”
The detection rate for cars has continually outstripped that for bicycles because improved techniques allow systems to train themselves by studying thousands of images where known objects are labeled. Not true with spotting and orienting bikes that make sudden turns or jump out of nowhere. That helps explain why performance plummets on bicycle detection. “Deep 3D Box is among the best, yet it spots only 74% of bikes in the benchmark tests. And although it can orient over 88% of cars, it scores just 59% for bikes.”
As U.S. regulators put it, “that means cyclists will be living a while longer with the human error that contributes to 94% of traffic crashes.” Q. Take a venomous snake, a mouse and a heroic biologist, and what good may come of it? A. You begin with a biologist willing to milk a poisonous snake of its venom, after which a nonlethal dose is injected into a horse or sheep for developing antivenomous substances that are used to create an antidote for humans, says Sean O’Neill in “New Scientist” magazine. However, a problem with this traditional method is that the numerous animal proteins in the antivenom can induce severe allergic reactions in people.
Enter U.K. research fellow Nick Casewell and colleagues, who are working to identify the 20 or so key toxins that cause life-threatening bleeding and will then immunize mice with them. “The resulting antibodies should combat the lethal effects of any venom from half of all snakebites.” Beyond this, plans are to clone the mice’s antibody-making cells to produce an endless supply of all-round antivenom—meaning animals would no longer be needed to produce it and the allergy issue would mostly be solved.
Reflecting on the risk involved in working so intimately with such lethal animals, Casewell concludes: “When I see first-hand the patients who are suffering, the destitution and poverty that these people live in and the effect that snakebite has on not only them but their families, it’s a powerful motivator.”