Analog Devices Inc. | Sponsor
(Image courtesy of Analog Devices Inc.)
Building a humanoid robot ranks among the most challenging tasks in modern robotics. On its own, such a system must coordinate movement, maintain balance, interpret visual input, and respond rapidly—all through an intricate network of joints, sensors, and data processors. The stakes rise even higher when the robot shares its workspace with people.
For humans, most of these actions feel effortless. We instinctively sense our center of gravity and shift it when running, jumping, or sidestepping an obstacle. We absorb enormous volumes of sound and sight at once, in real time, and instantly decide how to react—all within milliseconds.
A humanoid robot must replicate this level of environmental awareness using a suite of sensors, then analyze those signals fast enough to define a safe operating zone and take action to prevent harming anyone nearby.
“People are inherently unpredictable. Any robot designed to work alongside us must be prepared for that unpredictability,” explains Geir Ostrem, an Analog Devices Fellow in ADI’s Automotive Business Unit.
With labor shortages intensifying and more robots entering shared workspaces to boost productivity, a central challenge emerges: what does it take for a humanoid robot to work safely and effectively right next to people?
Vision
For humanoid robots, situational awareness begins with vision—particularly in settings where people and machinery are always in motion. To react swiftly and correctly, whether grabbing an object or stepping aside for a person, the robot must see and make sense of its environment. Standard human-like vision can be mimicked using RGB image sensors, while depth perception can be achieved through time-of-flight, structured light, or stereo vision techniques.
But simply capturing visual data isn’t sufficient; processing it rapidly and precisely is just as important. Cameras and visual sensors are typically positioned far from the central computer, so all that data must travel through the robot via lengthy cables, which adds delay.
In a humanoid robot, the main processor acts as the brain, with vision sensors mounted in the head or torso linked directly to this central unit. When faster response times are needed for tight control loops—such as driving a motor for quick movements—smaller auxiliary processors can be placed nearer to the sensor or actuator, handling local computation while also relaying data to the main processor.
Already widely adopted in the automotive sector, ADI’s Gigabit Multimedia Serial Link (GMSL) technology transmits video data in real time through a single stream that can carry multiple gigabits per second. In humanoid robots, this supports redundancy and rapid, on-board processing of visual information. This allows these systems to interpret what their surroundings look like using physical AI that processes visual data locally on the robot, rather than relying on the cloud.
Audio
Vision by itself isn’t enough, though. If a robot is going to collaborate with humans, it also needs smart hearing. “The ability to communicate in natural language through a conversational interface is an extremely effective way to interact with a robot,” says Ostrem.
Even more critical than a conversational interface, however, is the robot’s capacity to recognize acoustic events happening around it. If something falls to the floor behind the robot, it needs to pinpoint where the sound came from and understand what it signifies, Ostrem notes.
Much like visual data, audio signals must travel from multiple microphones to the robot’s central computer for analysis, making latency a key concern.
“For sound event localization and detection, having predictable, consistent latency from microphone to computer is absolutely essential,” Ostrem says. “You’re dealing with beamforming and the acoustic field, which demands precise knowledge of the timing differences between microphones.”
“That’s an area where ADI’s A2B audio bus excels—transmitting a signal from a microphone to a computer over A2B always takes exactly 63 microseconds.”
Another technology borrowed from the automotive world, A2B is a low-latency audio transport solution that supports sound source localization. It simplifies audio connectivity across the entire system by enabling multiple microphones to be daisy-chained on a single bus—delivering power, audio, and control signals over just two wires.
“When you examine a robot, the sheer volume of wiring required for all its sensors is one of the biggest headaches,” Ostrem says. “A2B lets you add sophisticated audio capabilities with remarkably few wires.”
Battery and Power
Every sensor, processor, and computing element needs energy to function. Humanoid robots aren’t plugged into a wall outlet, so they rely on integrated battery packs. Most humanoid robots operate on lithium-ion cells rated at 48, 60, or 72 volts—smaller than automotive batteries but carrying similar dangers, including overheating and thermal runaway.
ADI provides technologies like electrochemical impedance spectroscopy to catch unsafe shifts in battery chemistry at an early stage, allowing batteries to be replaced before they fail.
“This lets you peer deep into what’s happening inside the battery’s chemistry,” Ostrem explains. “If something goes wrong or the battery becomes unsafe in any way, you can spot it ahead of time—before it turns into a real problem. When a robot is operating near people, if that battery heads toward thermal runaway, you absolutely need to ensure the robot is well away from any human when the issue actually occurs.”
And all of this hinges on the audio, visual, and other internal sensors being able to detect issues, relay them quickly, and trigger actions that minimize safety risks to people nearby.
Conclusion
As humanoid robots take on increasingly complex responsibilities, the demands around safety, sensing, and human interaction will only intensify. Ostrem sees the future in stronger AI at the edge—both through more accurate object classification and more efficient low-power edge processing.
ADI has already mastered sensing and perception, connectivity, and battery management in the automotive domain. The logical next move is applying those proven technologies to emerging fields like humanoid robotics.
“In many ways, humanoid robots today are where automobiles were decades ago,” Ostrem says. “The architecture is still evolving, and there’s plenty of opportunity for industry collaboration on standardizing interfaces to grow the ecosystem around humanoid robots.”
For more information, visit analog.com/industrial-robotics/humanoid-robotics.
