Point One Navigation raises $35 million to turn centimeter-level positioning into infrastructure

Point One Navigation has raised a $35 million Series C led by Khosla Ventures. The pitch is pretty simple: location data for vehicles and robots should be accurate to centimeters, not meters, and most of that value should show up as software.

That sounds narrow until you look at where autonomy still fails. A robotaxi that knows it is in the lane but not exactly where it sits against the curb has a problem. So does a delivery fleet trying to prove where a truck stopped. In agriculture and turf care, 10 centimeters can be the difference between clean work and damaged property. The tolerances keep tightening, and plain GNSS isn’t close to enough.

Point One says its system can deliver position accuracy down to 1 centimeter in ideal conditions by combining RTK-corrected GNSS, computer vision, and sensor fusion. It already claims real deployment volume, including more than 150,000 vehicles, a 300,000-vehicle last-mile delivery fleet, a global motorcycle brand, and several mowing and turf-care manufacturers. The new money goes toward expanding its correction network, Polaris, across North America, Europe, and Asia, plus indoor positioning, where satellite fixes stop helping.

That last part matters. Plenty of companies can show precision on an open road in good weather. Keeping localization stable near tall buildings, under trees, through parking structures, and into industrial sites is where things get hard.

Three systems, one localization stack

Point One’s stack follows the pattern most serious autonomy teams already know. No single sensor gives you reliable localization everywhere. You combine several imperfect ones and try to keep failure modes manageable.

The first piece is augmented GNSS. Standard satellite positioning usually lands in the meter range, and it degrades fast in cities or under partial sky view. RTK, or real-time kinematic positioning, tightens that by using correction data from nearby base stations. Point One’s Polaris RTK Network reportedly uses lunchbox-sized stations spaced about every 40 kilometers. That spacing matters because RTK quality drops as the rover moves farther from the reference station and local atmospheric conditions stop lining up.

Those stations publish RTCM 3.x correction messages, usually over NTRIP, with the usual ingredients: carrier-phase corrections, ephemeris, ionospheric models. Pair that with multi-frequency, multi-constellation receivers across GPS, Galileo, BeiDou, and GLONASS, and centimeter-level fixes are possible when the environment cooperates.

Then there’s vision. Cameras help when GNSS gets unreliable. Lane markings, curbs, road edges, buildings, fixed landmarks, all of that can anchor a vehicle’s local position. In practice, this usually means VIO, visual-inertial odometry, plus map-based localization where available. Vision is good at relative motion, but it drifts over time and degrades in bad weather, poor lighting, occlusion, and dirty-lens conditions.

The third piece is the one that decides whether the product is usable: sensor fusion. GNSS, IMU, wheel speed, steering angle, and sometimes lidar have to be reconciled into a single pose estimate with uncertainty attached. Usually that means an Extended Kalman Filter, a factor graph, or some hybrid setup that can handle time synchronization, calibration, and error modeling without blowing up the compute budget.

This part gets undersold. High-accuracy localization isn’t just a pretty dot on a map. The system has to report covariance, detect bad inputs, recover from dropouts, and avoid sudden pose jumps that break downstream planning. For ADAS or autonomy features, integrity matters as much as raw accuracy.

A localization stack that’s usually right but fails without warning is dangerous.

Why the timing matters

The funding itself stands out. A $230 million post-money valuation for a positioning company suggests investors see precision localization moving from a specialist subsystem to a broader platform layer.

That shift has been building for years. RTK has long been standard in precision agriculture and surveying. What changed is deployment economics and vehicle architecture. Newer vehicles already carry some of the hardware you need: cameras, IMUs, wheel odometry, usable compute. If the GNSS receiver and antenna setup are good enough, the step up to better localization starts to look like software plus a network subscription, not a full custom hardware program.

That matters to OEMs and fleet operators. Hardware takes forever to certify, costs more to service, and is miserable to retrofit at scale. Software-first positioning is easier to sell, especially if the vendor can hide the ugly parts like station selection, correction transport, time sync, and calibration behind an API.

But software-first only works if the hardware underneath isn’t junk. A cheap antenna in a bad mounting position can ruin the whole stack. So can multipath, interference, poor IMU bias stability, or unsynced camera feeds. Localization has always been unforgiving.

RTK works. It also needs infrastructure.

Point One’s bet on RTK over PPP makes sense for vehicles, robots, and drones that need fast convergence and tight tolerances. PPP, or precise point positioning, gives you broader coverage and less dependence on dense local base stations, but it usually takes longer to converge and can be harder to trust for mobile systems that need precision immediately.

RTK gets there faster. The trade-off is obvious. You need a dense correction network, and dense networks are expensive to build and maintain.

That makes the Polaris expansion one of the more important parts of the story. If Point One wants to be a default positioning layer, coverage and continuity matter just as much as the fusion algorithms. A centimeter-grade positioning engine is only as good as the correction service behind it. Lose coverage in the wrong freight corridor or industrial zone and the product gets a lot less attractive.

The regional focus says something too. The Midwest and East Coast are sensible targets because they combine agriculture, freight routes, industrial activity, and dense road infrastructure. That’s a better place to make money than scattered coverage built for demos.

Indoor is the next hard problem

Point One says it wants stronger indoor capabilities, not just support for parking structures. That’s ambitious, and this is where things usually get messy.

Once GNSS fades, RTK is out of the picture. Indoor localization turns into some mix of SLAM, vision, inertial dead reckoning, UWB, BLE Angle of Arrival, Wi-Fi RTT, fiducials, or facility maps. Every option has trade-offs. Vision can work well in structured environments but struggles with repeated textures, poor lighting, and layout changes. UWB is reliable but requires anchors and deployment work. BLE is cheaper and less precise. Pure inertial dead reckoning drifts.

The real difficulty is the handoff. Outdoor absolute position and indoor relative localization have to transition cleanly. No jumps, no yaw flips, no sudden confidence spikes when the system is guessing. That’s harder than it sounds, especially for forklifts, yard robots, or delivery vehicles moving through mixed environments all day.

If Point One gets that right, the product starts to look like real operational infrastructure for industrial robotics. If not, the indoor story stays aspirational.

What technical teams should look at

If you’re evaluating Point One or anything similar, ignore the 1 cm headline for a minute and ask harder questions.

Sensor and hardware requirements

Software can hide complexity, but it can’t repair bad inputs. You still need:

• dual- or tri-frequency GNSS with carrier-phase support
• good antenna placement and RF design
• an IMU with known bias behavior
• wheel speed and steering data when available
• camera and IMU synchronization for VIO

Without that, the fusion layer spends its time compensating for hardware mistakes.

Time sync and uncertainty reporting

Any vendor can stream pose estimates. The useful question is whether those estimates are timestamped properly and whether they include usable covariance or integrity metrics. If your planner, safety stack, or map alignment logic can’t reason about uncertainty, the localization pipeline is incomplete.

Failure detection

Spoofing, multipath, camera dropout, IMU drift, station outages, all of this happens in production. Ask how the system flags degraded states and what fallback behavior looks like. RAIM-style consistency checks and cross-sensor validation matter more than impressive median accuracy.

Latency and compute budgets

Fusion pipelines look great in offline evaluation and a lot less friendly in production. Vision stacks eat compute. High-rate inertial updates need careful scheduling. Correction transport depends on the network. On constrained embedded systems, every localization gain comes with a power and latency cost.

Integration model

Point One’s API pitch is attractive if you want a clean abstraction over corrections and localization. It’s less attractive if you need raw access for custom estimators, safety certification work, or deep integration with an existing autonomy stack. The more opinionated the platform, the faster adoption tends to be and the tighter the coupling becomes.

A crowded market, but the layer makes sense

This isn’t an empty category. Swift Navigation, Trimble, Hexagon, u-blox, and plenty of OEM internal teams all work on parts of this stack. Point One’s angle is vertical integration without a big custom hardware ask up front: base stations, correction network, fusion software, API surface. That’s a sensible place to compete.

It also means the company has to be good at two very different businesses: infrastructure operations and estimation software. A lot of startups can do one. Fewer can do both well.

The funding round matters less than the broader shift behind it. Precision localization is becoming a standard dependency for autonomy, ADAS, robotics, and industrial fleets. Once teams need lane-level confidence, curb-level stopping, or repeatable centimeter paths, “good enough GPS” stops being good enough.

Agriculture crossed that line years ago. Other sectors are now catching up.