Robots need more affordable sensors and low-power processing to enhance safety and gain widespread adoption in industry and consumer settings, Texas Instruments engineers said Thursday.
“Sensing is a focus for us,” said Sameer Wasson, vice president of processors at TI in an online roundtable with analysts and reporters. “There is no one sensor for every everything…the sensing suite depends on the app and multi-modality is probably a fundamental requirement.”
Sensors needed for robots include simple cameras, magnetic sensors, LiDARs and radar, he added.
“Cost and affordability [of sensors] is equally important for robots to reach mainstream,” he added. “That’s where the art is and it’s an interesting place for engineers.”
Sensors are considered important for helping robots work autonomously at the network edge, in applications where a constant connection to the cloud isn’t helpful. In one example, Wasson described how a robot lawnmower needs to constantly adjust power to climb a bumpy hill, something requiring sensors and instant processing.
By contrast, in logistics and last-mile delivery applications, robots will need to rely on machine learning and real time communications—sometimes with other robots—to navigate through a factory to pack, pick, sort and carry goods. Wasson said there isn’t yet a reliable robot-to-robot communication standard although there are sub 1 GHz wireless standards that are starting to permeate the industry.
Affordable and efficient sensing is needed for safety as well. Sensing is needed to adapt to the presence of any humans “to make sure the robot is going to stop,” added Giovanni Campanella, systems manager for robotics at TI.
“We don’t see as many robots as we’d like,” Campanella added. “One reason is expensive modules, sensors and cameras.”
Wasson said a focus on safety will improve robot adoption, especially in consumer settings. “If you can’t give confidence o the end use that this [robot] is safe, you won’t see continued adoption.”
Robot vacuum cleaners, for example, need better sensing to know whether they are approaching a liquid spill on the floor or have reached the edge of stairways. “Robotic vacuums now don’t really see what they are picking up…They don’t see staircases well and we don’t want them tumbling down. Things like that prevent the technology from becoming mainstream.”
Increasing the overall autonomy of robots would improve robot adoption, Campanella added. “Increasing autonomy would really change the game and you increase that with two things---sensing and processing. Precise and accurate sensors are important to tell short and long distances and which objects are in front of it to make certain decisions if a human approaches….Enabling autonomy intelligence is one enabler to seeing more robots around us.”
Wasson said it is hard to put a timeline on greater adoption of robots in homes and industry. “Every year we do make progress. Anything safety-related is going to take time,” he said.
TI designs a broad range of processors that are used by robots in multiple applications, competing with Qualcomm, Intel, Nvidia and many others. “Our unique value proposition is our capacity to do this in a constrained power environment,” Wasson said. “If you put a supercomputer in a robot, it will have processing power [to do many things] but will work for 10 minutes and run out. We focus on how to do machine learning in a power constrained environment… with a wider portfolio for connectivity and drives.”
TI recently posted an e-book with technical documentation on robotic system designs.