Last month, Caltech opened a new interdisciplinary center to study and develop the robots of the future – drones and machines equipped with artificial intelligence that will allow them to make their own decisions and partner with humans on sci-fi sounding projects like a fully automated flying ambulance and a bipedal robot that can help scientists explore Mars. That’s the dream, anyway.
AirTalk fill-in host John Rabe and Sanden Totten, co-creator of kids' science podcast “Brains On!” and a science writer for “Bill Nye Saves the World,” went inside the Center for Autonomous Systems and Technologies to see what’s being developed and talk to the researchers about their goals, as well as the limitations and ethics of autonomous robots.
Here are some of the most interesting things we learned and saw.
This three-story tall drone arena is equipped with nearly 1,400 fans that generate winds that go up to 45 miles per hour.
What’s it for? The fans simulate various wind and storm conditions so researchers can test how drones react in real-world weather. The aerodrome is also rimmed with 48 cameras that follow the motion of the drone.
Are the drones fully autonomous? Not yet. But according to Mory Gharib, director of CAST and head of the aerospace department at Caltech, the cameras capture information about how the drone is reacting to remote-controlled commands as it struggles against simulated winds. This information is used to develop the drone's software so it can learn "the ability to reason or negotiate the wind" and eventually become fully autonomous.
The flying ambulance
What’s it for? Rescuing people from hard-to-reach spaces and transporting them quickly to nearby hospitals, while monitoring some of their vitals. Its stretcher would be automated as well.
Fun fact: The design was based on a boxfish, whose shape is optimized to withstand water flow so it can maneuver and stabilize itself in currents.
The frictionless floor
This floor in the Space Robotics Lab operates like an air hockey table, with jets shooting a thin cushion of air that satellite simulators glide on.
What’s it for? The goal of the lab is to be able to send automated drones into space and have them communicate and coordinate with each other to build structures, such as a telescope. The frictionless floor allows researchers to simulate space-like conditions.
Fun fact: It takes a very special robotic vacuum to clean this floor without making it dirtier.
What's it for? The prosthetic leg is powered, which means its wearer can expend less energy when walking. The goal is to have the prosthetic leg adapt to its user’s gait and help them be more symmetric and comfortable.
So what's going on inside the prosthetic leg? "It is actively looking at [the user's]...movements to decide how to move. So, every moment in time we have mathematical expressions that are evaluating on the device and deciding what to do next," said Aaron Ames, professor of mechanical engineering and control and dynamical systems.
Cassie, the bipedal robot
What’s the point of a bipedal robot? Wouldn’t a quadrupedal robot be more stable? Aaron Ames said bipedal robots have the capacity to explore certain areas that robots with other designs cannot reach. The goal is to one day send Cassie to Mars where it would have the capacity to explore spaces that a rover cannot. Also, since our world was created for bipedal creatures (read: us), this type of robot will be able to navigate the human world more easily.
What’s the short-term goal for Cassie? Currently, CAST is working on getting Cassie to hike the Pacific Crest Trail from Mexico to Canada. The goal is for Cassie to be fully autonomous and capable of sensing and adjusting to different kinds of terrain.
Fun fact: Cassie and other bipedal robots fall over a lot, which is why Caltech's Advanced Mobility Lab is padded with gym mats.
We also talked to Anima Anandkumar, a machine learning expert who describes her job as putting the brains inside the robots, about some of our concerns regarding AI.
Will the robots become smart enough to take over?
What is intelligence? How adaptive can the AI systems be? Is it able to understand the underlying meaning of the tasks? Is it able to then learn better, teach itself new skills? We are still very far from that.
Is the concern about automation taking jobs justified?
That part is justified, because many industries are ripe for disruption and this is where the social aspect and having the right support systems makes a huge difference, because if done wrong then in the short term you will see many people out of work.
So what concerns you most about an AI- and drone-filled future?
Anandkumar said she is most concerned about AI reinforcing and multiplying the biases of humans.
The primary concern is that we should be aiming to democratize AI. It should not be only accessible to few... The other aspect is the kind of data we feed into the system, because the biases that can build up [in people and society] get multiplied several times more [in AI].
What do we do about it?
We come up with principled approaches to data collection. What does it mean to collect balanced data? Can we have fairness measures for our algorithms and can we prove that they satisfy these requirements?