On July 29, it was reported that Beawkward recently released the NOVA20 solid-state array near-field radar.—Adding safety perception to the front 5cm–1m of the embodied robot:It won't cause blindness from sun exposure, doesn't consume much computing power, can be distributed around the body, and provides deterministic input for secure redundancy..
The hottest buzzword for embodied robots this year is "brain"—VLA, world model, data, generalization. But the stronger this brain, the better...The more computing power is consumed (computing power equals cost, which equals electricity), the more it consumes.,returnOccasionally I might misread or misunderstand.—And it isThousands of unitsWalk up to the person.
When robots are truly entering factories, streets, and homes, a more fundamental question must be addressed first: how to ensure their...Without bumping into anyone, without falling down the stairs, and without going blind in the sun.The "close-fitting safety" layer 5cm-1m in front of you is a gap that cannot be covered by 360° rotating radar, head cameras, or ultrasound.And it,We shouldn't have gambled on a brain that's prone to making mistakes..
North Awakening NOVA20 perfectly fills this gap, adding another layer to the robot.It doesn't consume computing power and can be stopped at any time.A close-range defensive line.
01.
Entering a crowd requires ensuring safety within the immediate vicinity of that one-meter radius.
The intelligence of embodied robots is advancing rapidly. However, before they can be truly deployed alongside people and operated on a large scale, there is one more fundamental thing that must be done thoroughly—5cm to 1 meter in front of the bodySafety on that circle: Don't crash, don't fall, don't have an accident.
This close-range, near-field situation precisely fills a gap in existing sensing capabilities:360° rotating radar(Blindness within a few tens of centimeters)Head camera(Blocked by the robot's own limbs, and also by the minimum distance)Ultrasonic/Infrared(Too few measurement points and large blind spots) None of the three are covered.
Moreover, this is not a niche issue: many cutting-edge...VLA/Basic Robot ModelResearch on this topic still primarily relies on vision and proprioception, with near-field safety typically left to the underlying system of the entire machine. Some leading teams (such as Gemini Robotics) have explicitly stated that "life-saving tasks" like obstacle avoidance and contact force control are better suited to be handled by...Independent of high-level models, deterministic low-level control—For personal safety, it is not advisable to rely solely on high-level models that are prone to making mistakes.
The NOVA20 was designed to fill that gap.It doesn't replace the brain, nor does it handle the fine motor skills of the hands (grasping, fitting, and force control are the domain of wrist cameras and haptics)—it does something else equally crucial: enabling a robot that can already perform tasks to reach people more safely and on a larger scale.
02.
Where is the gap: under your feet, beside you, and among the people around you?
NOVA20 supplements those...The camera didn't cover it at all.The location isn't "poorly readable," but rather "lacking in sensors." Three locations are the most typical:
1. Footprint - Prevent falls
▲The overhead camera can't see what's underfoot; it misses steps, drops, and cliffs, and one wrong step can lead to a fall. NOVA20 uses 20 points to simultaneously monitor slope edges and drops, detecting them earlier than single-point infrared.
Its official purpose is to prevent falls from your feet.Sensors mounted high up and looking forward can be blocked by the device itself, causing the device to "only realize it's close to the cliff and not have time to stop" (ETH ArtPlanner, LimX, and Shanghai Stock Exchange all have evidence of this); misstepping on open steps and hitting the edge while going downstairs are known points of failure.
2. Both sides of the body · The outer perimeter of the whole body
▲When turning or swinging its arm, the robot will use the limb without sensors to collide with people or objects to the side. By attaching NOVA20 to the limb, that limb can stop itself before making contact.
The camera is often positioned at the far end, with a limited field of view, making it difficult to see the torso and the area around the limbs.It avoided the section with the sensor, but instead used the section without the sensor to hit the person..
Apple and CMU's ARMOR research40 Time-of-Flight (ToF) sensors are distributed along the humanoid arms.This allows the robot to see things up close from its own perspective—a perspective far superior to relying solely on a camera on its head.The collision rate decreased by 63.7%.He also pointed out that "centralized cameras can be blocked by one's own arms and hands."Distributed along the body is better than concentrated on a single camera.This is exactly the idea behind NOVA20.
3. There are people around; strong outdoor light
▲Relying on "relief upon contact" means only taking remedial action after a collision; NOVA20 discovered before contact that it slows down or stops as soon as a person gets close.
Moreover, this layerEspecially blank outdoorsIn sunlight, indoor close-range vision systems are essentially ineffective. Accidents involving robots near humans are no longer hypothetical—they have already occurred within the industry.Delivery robots injure pedestrians outdoors; humanoid robots collide with workers during testing.Multiple close-range safety incidents have resulted in real injuries and liability. Furthermore, the 360° rotating radar is large and heavy, making it difficult to deploy points along the entire fuselage for close-range observation.
The commonalities among these three places:The camera can't capture it at all; it's because there's "no sensor," not because the sensor is faulty.This is precisely where the NOVA20 focuses its efforts. (Fine grip is the domain of wrist-based cameras and haptics, which the NOVA20 doesn't address.)
03.
True value:
Provide deterministic input for near-field safety redundancy of the entire system.
For embodied robots to be deployed on a large scale in human-centric environments, safety should not rely solely on VLA or visual models.NOVA20 provides a near-field ranging input independent of the high-level model.Low power consumption, no image output, strong light resistance, and can be deployed along the body, helping the whole device to...Underfoot, beside, and next toBuild verifiable security redundancy.
▲A solid-state multi-point near-field radar—20 points spread out in a 40° fan shape, directly measuring distances from 5cm to 1m; no rotation, no moving parts, no image output.
Why is this matter particularly crucial in the "mass production" stage?Because large-scale deployment requires safety case verification, and such verification needs a deterministic bypass that does not rely on a large model.It's difficult to prove that "a probabilistic large-scale visual model will never bump into a person"—it has high latency jitter, consumes a lot of computing power, and will go blind if it crashes.
And NOVA20:Less than 1 watt, no image output, 100 Klux glare resistance, deterministic triggeringJust for thisSafety redundancy that does not rely on probability and can independently stop in an emergency.Provide input.
No matter how powerful a camera or large model is, it cannot replace this channel—security verification requires deterministic input that is "not based on probabilistic models," which is precisely what cameras and large models cannot naturally provide.
Relevant standards are also moving in this direction:The revised version of the industrial robot safety standard ISO 10218-1:2025, to be released in 2025.It incorporates a safety capability framework for collaborative applications such as "Speed and Distance Monitoring (SSM)"—that is, when the robot approaches a person, it uses reliable sensors to measure the distance in real time and slows down accordingly. It is primarily aimed at industrial robots; however, this approach of "maintaining human-robot distance through deterministic ranging" is precisely what NOVA20 offers.Potential key inputs.
OrientedISO 25785-1 Safety standard for dynamically stabilized mobile robots (including some humanoid models) in industrial environments.It is also under development (currently in the Committee Draft stage and has not yet been officially released; the drafting group is led by Agility Robotics and includes Boston Dynamics, etc.).
It should be noted that its application scope is industrial environments, and does not mean it covers all embodied or home robots. However, as such standards gradually take shape, "how to verify that it will not collide with people" will become increasingly important—and a near-field input layer that does not rely on probability models and can be triggered independently is precisely one of the necessary components.
04.
Why was it Bei Xing who made this?
NOVA20's attribute combination—<5g, <1W, no image output, controllable cost, resistant to 100Klux, can be distributed along the body.—It is a hard currency that can be mass-produced.
This confidence comes from Beixing.Nearly nine years of dedicated work on the path of "pure solid-state obstacle avoidance LiDAR".From the first-generation pure solid-state area array LiDAR CE30 in 2017, to the first-generation solid-state small-angle linear LiDAR NOVA in 2024, and now to the pure solid-state array LiDAR NOVA20 released today—each generation has resulted in smaller size and lower cost.Finally, the long-standing problem of "crosstalk" has been completely solved..
It is this accumulation of knowledge and experience over ten years, starting from obstacle avoidance, that has enabled Beawkward to make this near-field safety layer a near-field safety component that robots can truly afford to install and that can be purchased on a large scale.
▲One sensor covers a 40° area nearby, such as the wrist, chest, back, waist, knees, and ankles. You can attach a few sensors wherever you want to protect the robot, creating a ring of sensors around the robot itself, without needing to stack large and expensive LiDAR sensors all over its body.
NOVA20 Press"Deployment based on demand"Design—whether it protects the feet, sides, or the whole body, you can simply apply the stickers to the areas you want to protect, and it comes with a complete integrated solution.Pure solid-state + 100Klux resistance + no image output (authentication/privacy friendly) + cost-controllable + can be distributed along the bodyProtecting the feet, sides, and entire body—this combination, designed from the ground up for robots, is currently...There are still few mature, mass-producible near-field safety components.What Beixing wants to do is to make it something that can be purchased on a large scale and that people can afford.
05.
Conclusion: Giving robots a ring of eyes that can "see clearly what's around them"
▲Each performs its own function—<5cm is handled by touch, 5cm–1m by NOVA20, and >1m by the 360° rotating radar and camera.
A brain makes robots smarter, a 360° rotating radar allows them to see into the distance, and NOVA20 makes it...Every arm, every leg, every stepAll before contactNo collisions, no fallsPut oneRobots that are already capable of performing tasks can be deployed more safely and on a large scale around people.—This near-field safety layer has been developed by Bewow into a near-field safety component that can be distributed along the body and mass-produced.