Inside a brightly decorated lab at the University of California, Berkeley, an ordinary-looking robot has developed an exceptional knack for picking up awkward and unusual objects. What's stunning, though, is that the robot got so good at grasping by working with virtual objects. The robot learned what kind of grip should work for different items by studying a vast data set of 3-D shapes and suitable grasps. The UC Berkeley researchers fed images to a large deep-learning neural network connected to an off-the-shelf 3-D sensor and a standard robot arm. When a new object is placed in front of it, the robot's deep-learning system quickly figures out what grasp the arm should use.

UC Berkeley researchers are tapping the cognitive smarts of babies, toddlers and preschoolers to program computers to think more like humans. "Children are the greatest learning machines in the universe. Imagine if computers could learn as much and as quickly as they do," said Alison Gopnik a developmental psychologist at UC Berkeley and author of "The Scientist in the Crib" and "The Philosophical Baby." In a wide range of experiments involving lollipops, flashing and spinning toys, and music makers, among other props, UC Berkeley researchers are finding that children -- at younger and younger ages -- are testing hypotheses, detecting statistical patterns and drawing conclusions while constantly adapting to changes. "Young children are capable of solving problems that still pose a challenge for computers, such as learning languages and figuring out causal relationships," said Tom Griffiths, director of UC Berkeley's Computational Cognitive Science Lab. "We are hoping to make computers smarter by making them a little more like children."

BERKELEY – When scale model helicopters pass through a makeshift "urban canyon" in a test field, or engage in a game of aerial "chicken", the drills may look like a robotic stunt show to outside eyes. Members of the university's Berkeley Aerial Robot (BEAR) program have successfully conducted a series of field tests with 130-pound helicopters that not only fly autonomously -- without human control -- but that also react to avoid obstacles in their flight path. "Our BEAR group is the first to successfully develop a system where autonomous helicopters can detect obstacles, stationary or moving, and recompute their course in real-time to reach the original target destination," said David Hyunchul Shim, a research engineer on the project who first began this work as a UC Berkeley Ph.D. student in mechanical engineering. With these achievements, the researchers are inching towards a future of robo-copters that could maneuver through city streets or forested landscapes. The development of reliable systems that can handle obstacle-avoidance tasks is still several years away, researchers said, but the computational foundations for such unmanned aerial vehicles (UAVs) have been laid.

Meet SALTO: a powerful new wall-jumping robot built by researchers at UC Berkeley. According to SALTO's makers, the diminutive, one-legged hopper not only has the "highest robotic vertical jumping agility ever recorded," but also the ability to link together multiple jumps in quick succession. SALTO stands for saltatorial locomotion on terrain obstacles, and the motion of the mechanical jumping leg was modeled after galagos -- small jumping primates native to Africa that have stretchy tendons in their legs that allow them to store energy and jump with more force than if they only used their leg muscles alone. The galago is so agile not only because it can make a big leap, but also because it can essentially wind up its legs into a crouched position in mid-flight and immediately leap again upon landing. At just 100 grams and 26 centimeters (10.2 inches) tall when fully extended, SALTO can jump a little bit more than one meter (3.3 feet) high in a single leap.

New'deep learning' technique enables robot mastery of skills via trial and error. UC Berkeley researchers have developed algorithms that enable robots to learn motor tasks through trial and error using a process that more closely approximates the way humans learn, marking a major milestone in the field of artificial intelligence. They demonstrated their technique, a type of reinforcement learning, by having a robot complete various tasks -- putting a clothes hanger on a rack, assembling a toy plane, screwing a cap on a water bottle, and more -- without pre-programmed details about its surroundings. "What we're reporting on here is a new approach to empowering a robot to learn," said Professor Pieter Abbeel of UC Berkeley's Department of Electrical Engineering and Computer Sciences. "The key is that when a robot is faced with something new, we won't have to reprogram it. The exact same software, which encodes how the robot can learn, was used to allow the robot to learn all the different tasks we gave it."