The next phase of machine vision innovation is unlikely to come from incremental improvements alone. It will come from rethinking long-standing assumptions, particularly around cost, performance, and system design.
That is where startups tend to matter most.
In the second part of our Startups to Watch series, we look at Efference, a company focused on one of the more persistent challenges in vision systems: reliable, affordable depth perception.
Rethinking Stereo Vision
Stereo cameras have long been positioned as a practical alternative to more expensive sensing technologies. In reality, they remain difficult to calibrate, sensitive to environmental conditions, and often inconsistent in real-world deployment.
At the same time, the cost of deploying vision systems, particularly in robotics and autonomy, remains high. Sensor stacks alone can push system costs into five figures, slowing adoption at scale.
Efference’s core thesis is simple: depth perception should not be this expensive or this unreliable.
From Biology to Vision Systems
Founder Gianluca Bencomo has built his career at the intersection of biology, neuroscience, and machine learning.
His background spans work at NASA’s Jet Propulsion Laboratory, research at Harvard Medical School on visual perception, and studies at Princeton focused on biologically inspired machine learning models.
That trajectory informs Efference’s approach.
Rather than treating depth as a purely hardware problem, the company is applying principles drawn from human vision, combining geometry with learned priors to interpret spatial information more effectively.
Depth as a Software Problem
Efference’s key idea is to shift the burden of depth estimation away from hardware and into software.
Instead of relying solely on stereo camera geometry, the company layers data-driven perception models on top of existing sensors. These models generate dense depth maps in real time, with reduced noise and added confidence estimates.
This approach has several implications:
- Existing cameras can be enhanced rather than replaced
- Depth estimation becomes less dependent on precise calibration
- Systems can operate with lower latency and reduced compute requirements
The company is developing this across three areas:
- An API to improve existing camera systems
- On-device models for real-time processing
- A proprietary camera platform
All processing is designed to run locally, removing the need for external GPUs and enabling deployment in edge environments.
Why This Matters Now
Efference is entering a crowded space. By some estimates, more than 100 companies are working on depth sensing and stereo vision.
The question is not whether the problem exists. It clearly does.
The question is whether software-led approaches can meaningfully close the gap.
If successful, this model would align with a broader shift already visible across machine vision: moving from hardware differentiation toward software and system-level intelligence.
It also reflects a growing trend toward reducing sensor complexity in favour of more capable processing pipelines.
A Promising, but Unproven Approach
There are reasons to be cautious.
Depth estimation remains highly sensitive to real-world conditions, including lighting, texture, and motion. While biologically inspired models are compelling, translating them into robust industrial performance is not straightforward.
Efference also faces competition from:
- established stereo solutions
- depth cameras
- LiDAR systems, which continue to fall in cost
And like many early-stage companies, it has yet to publicly demonstrate large-scale deployment or customer adoption.
What Comes Next
Efference has recently completed the Y Combinator program and is actively building out its team.
Its ability to scale will depend less on technical promise and more on execution:
- securing OEM partnerships
- proving reliability in real-world environments
- integrating into existing vision stacks
If it can do that, its positioning is clear.
A system that delivers reliable depth without the cost and complexity of current approaches would have immediate relevance across robotics, drones, and automotive systems.
A Signal of a Broader Shift
Whether Efference succeeds or not, the direction is telling.
The industry is increasingly treating perception not as a fixed capability of hardware, but as something that can be improved, extended, and redefined through software.
That shift is already visible across AI-driven imaging pipelines, simulation environments, and edge deployment strategies.
Efference is one example of how that thinking is being applied to depth.
And for that reason alone, it is a startup worth watching.
