Integrated Navigation for Low-Speed Unmanned Vehicles
· ⏱ 10 min read · 👁 viewsShare
As autonomous driving moves from trials into real-world operation, low-speed unmanned vehicles are becoming an important part of smart logistics. Compared with passenger vehicles on public roads, they travel more slowly and usually follow fixed routes. This makes it practical to deploy them earlier in industrial parks, campuses, tourist sites, airports, mines, and other controlled areas, particularly for last-mile delivery.
Low speed and fixed routes, however, do not make positioning easy. Buildings, trees, underground garages, tunnels, and large loading-area equipment can weaken or completely block GNSS signals. The vehicle still needs continuous position, velocity, and attitude information for planning, control, and safe stopping. The M20X dual-antenna integrated navigation receiver combines a K922 high-precision positioning and heading board, an automotive-grade six-axis IMU, RTK and integrated navigation algorithms, and online calibration to provide stable high-precision pose outputs for low-speed unmanned vehicles.
1. Application Prospects of Low-Speed Unmanned Vehicles
Low-speed unmanned vehicles are increasingly being deployed in controlled environments where repetitive tasks and defined operating areas make autonomous operation practical.
Typical applications include:
- Smart logistics: Last-mile delivery, warehouse transportation, and parcel distribution
- Passenger transportation: Autonomous shuttle services in campuses, industrial parks, and tourist sites
- Mining: Unmanned transportation and material handling
- Inspection: Autonomous inspection of industrial facilities and infrastructure
- Cleaning: Unmanned cleaning vehicles for roads and public areas
- Food and retail: Autonomous food delivery and unmanned retail services

Figure 1. Common application scenarios for low-speed unmanned vehicles.
These vehicles commonly operate on urban side roads or within industrial parks, campuses, tourist sites, airports, and mines. Because routes are fixed and the environment changes less than on public roads, perception and control algorithms can be made more robust, helping autonomous operation reach commercial use sooner within restricted areas.
Demand is also growing rapidly. By the end of 2024, China's express-delivery volume had reached 175.08 billion parcels, up 21.5% year on year, while the number of online food-delivery users reached 590 million, an increase of 8.8%. In May 2025, China's Ministry of Commerce and seven other departments issued the Special Action Plan for Accelerating the Development of Digital and Intelligent Supply Chains. The plan calls for lower logistics costs and wider use of smart warehousing, autonomous tractors, unmanned delivery, and intelligent coordination of people, vehicles, and goods.
Pilot programs in Shenzhen's Pingshan industrial area involving JD Logistics, Meituan, and SF Express have introduced all-weather unmanned delivery operations. The pilots handle about 80,000 deliveries per day, reduce the average cost per order by RMB 1.32, and improve delivery efficiency by roughly 30%, while easing labor shortages during night shifts. For operators, the value extends beyond replacing a driver: vehicles can run more frequently, accident exposure can be reduced, and dispatching becomes easier to quantify and optimize.
2. GNSS/INS Integrated Navigation for Low-Speed Unmanned Vehicles
Passenger-vehicle autonomous driving often relies heavily on wide-area HD maps and end-to-end perception. Low-speed unmanned vehicles operate in smaller areas, where a more common approach is to combine lidar, cameras, and a local high-precision map. The vehicle can build the local map while operating, and high-precision GNSS ties the local coordinate frame to a global reference frame.
Figure 2. The M20X dual-antenna integrated navigation receiver has been deployed in more than ten mass-produced vehicle models.
Every individual sensor has limits. Rain, fog, dust, changing illumination, and repetitive environments can cause lidar or visual matching to fail. Urban canyons, tree cover, and low antenna installation heights can introduce GNSS multipath, degrade observation quality, or cause a complete outage. Whether the vehicle is building a map or driving autonomously, continuous and reliable high-precision pose information is fundamental to Level 4 operation.
The M20X incorporates the K922 high-precision positioning and heading board. In open sky, RTK provides centimeter-level positioning. The dual-antenna baseline also provides heading in real time, allowing heading initialization while stationary and continuous heading constraints at low speed.
When the vehicle enters an underground garage or tunnel, GNSS can no longer provide valid observations. The six-axis IMU measures angular rate and specific force to propagate position, velocity, and attitude. If wheel speed is available, the vehicle's forward velocity can further constrain inertial drift. Once GNSS returns, the satellite solution corrects IMU bias, attitude, and position errors. The purpose of integrated navigation is to keep the output continuous as signals move from normal reception to degradation, complete outage, and recovery.

Figure 3. GNSS, IMU, and wheel-speed observations complement one another, allowing the navigation filter to maintain position, velocity, and attitude through normal reception, signal outage, and recovery.
3. Measurement and Estimation of Integrated Navigation Extrinsics
After the receiver is installed, lever arms must be measured and configured accurately, and the mounting angles must be calibrated. These extrinsic parameters are not merely installation records: they are used directly in observation transformations and state estimation. Incorrect values can cause the filter to interpret installation offsets as vehicle motion.
GNSS Constraints and the Antenna Phase-Center Lever Arm
The GNSS antenna is usually mounted on the roof, while the integrated navigation receiver is installed near the chassis or inside an equipment bay. The vector from the receiver-frame origin to the primary antenna phase center, expressed in the vehicle frame, is the antenna phase-center lever arm. Because the two points are spatially separated, the GNSS antenna position and velocity must be transformed to the receiver origin:

Here, p₍b₎ⁿ and v₍b₎ⁿ are the position and velocity of the receiver origin in the navigation frame; lᵛ₍bg₎ is the lever arm from the receiver origin to the antenna phase center; and v₍b₎ⁿ and c₍b₎ⁿ describe the attitude transformations among the vehicle, receiver, and navigation frames. Vehicle rotation also produces additional linear velocity at the antenna, so lever-arm errors become particularly visible during turns and attitude changes. The lever arm should be measured as accurately as possible, or estimated by online calibration when direct measurement is impractical.
Vehicle-Motion Constraints, Rear-Axle Lever Arm, and Mounting Angles
Vehicle-motion constraints are normally defined at the center of the rear axle, whereas the receiver is rarely installed at that point. The vector from the receiver origin to the rear-axle center is the rear-axle lever arm. The receiver axes may also be misaligned with the vehicle's forward, right, and down axes, so the mounting angles must be applied using a Z-Y-X rotation sequence. The velocity at the rear-axle center can be written as:

For a front-steered Ackermann vehicle moving normally with a small steering angle, the lateral and vertical velocities at the rear-axle center can be approximated as zero. If wheel speed is available, the forward velocity at the rear axle can be approximated by the average of the left and right wheel speeds. These constraints can substantially improve velocity and position propagation during a GNSS outage.
The rear-axle lever arm should be measured manually. Mounting angles may be measured or obtained through online calibration. Ideally, the receiver should be mounted level, with its X and Y axes aligned with the longitudinal and lateral axes of the vehicle. M20X online calibration can estimate the antenna lever arm and receiver mounting angles in about five minutes.
Performance in an Underground Garage
To evaluate the effects of extrinsic calibration and wheel-speed aiding on outage propagation, a SPAN-ISA-100C fiber-optic integrated navigation system was used as the reference. The vehicle experienced a continuous GNSS outage of 281 s while traveling about 1.14 km through an underground garage. Its maximum speed was 7.31 m/s and its average speed was 4.07 m/s. Three configurations were tested: inaccurate extrinsics, calibrated extrinsics, and calibrated extrinsics with wheel-speed aiding.

Figure 4. M20X three-dimensional position error during the GNSS outage.
| Simulation setup | Outage distance (m) | Maximum error (m) | Drift rate | Position RMSE (m) |
|---|---|---|---|---|
| Test 1: Incorrect parameters, no wheel speed | 1140.79 | 11.973 | 1.05% | 7.086 |
| Test 2: Calibrated, no wheel speed | 1140.79 | 1.799 | 0.16% | 0.609 |
| Test 3: Calibrated, wheel-speed aiding | 1140.79 | 0.928 | 0.08% | 0.493 |
Table 1. Summary of M20X GNSS-outage simulation results.
In Test 1, both the antenna and rear-axle lever arms contained 30 cm errors; the mounting angles were not estimated and wheel speed was not used. The maximum position error reached 11.973 m, with a drift rate of 1.05%. After the extrinsics and mounting angles were correctly calibrated, the maximum error fell to 1.799 m and the position RMSE to 0.609 m. With wheel-speed aiding added, the maximum error decreased further to 0.928 m and the drift rate to 0.08%.
Figure 5. Test trajectory through the underground garage.
The results show that outage performance is not determined by IMU grade alone. When lever arms or mounting angles are inaccurate, every turn and attitude change can feed installation errors into the navigation filter. Correct calibration reduced the maximum error by about 85%. Adding a forward-speed constraint from the vehicle reduced it by a further 48%. For platforms that must operate through long GNSS outages, reliable chassis wheel speed should therefore be integrated whenever possible.
4. Application Challenges for Low-Speed Unmanned Vehicles
Challenges in Low-Speed Operating Environments
At low speed, a single-antenna GNSS receiver takes longer to obtain a stable course-over-ground heading. Antennas are also commonly mounted low on the platform, where the body, buildings, and vegetation can cause more obstruction. Many compact chassis lack passenger-car suspension, so ramps, curbs, and rough surfaces transmit stronger vibration to the receiver. These factors affect initialization, GNSS observation quality, and dead reckoning during signal outages.
The M20X uses dual-antenna heading to initialize while stationary and to constrain heading during low-speed motion. It also includes a GNSS quality-control module that screens satellite observations before they enter the navigation filter. Its inertial algorithm is optimized for high-vibration conditions to improve propagation stability in obstructed environments.
Figure 6. Examples of two-wheel differential-drive unmanned vehicles.
Challenges Posed by Different Vehicle Models
Low-speed unmanned vehicles use several chassis architectures. A two-wheel differential-drive platform has two independently driven wheels and one or two caster wheels, allowing it to turn in place. A tracked platform can also be treated as a generalized two-wheel differential-drive model. The kinematics are relatively simple, and wheel-speed constraints are usually straightforward.
Four-wheel differential-drive platforms have more complex dynamics and kinematics and are commonly used for outdoor transport, defense, or exploration. During a turn, the lateral and vertical velocities at the rear-axle center may not be close to zero. Applying the nonholonomic constraints of a two-wheel differential or Ackermann model without modification can therefore introduce incorrect observations into the filter.
Multi-wheel-steering vehicles offer a smaller turning radius and a wider range of maneuvers than conventional front-steered Ackermann platforms, making them common in AGVs. During coordinated multi-wheel steering or special maneuvers, the rear axle may have significant lateral motion, and wheel speed cannot be used directly as a simple forward-velocity measurement. The integrated navigation receiver must select its constraint model according to the number of driven wheels, steering architecture, and current chassis state. The M20X supports differential drive, multi-wheel steering, and other special operating modes for different low-speed unmanned platforms.
Receiver selection is only the first step in deployment. Engineers should define the coordinate axes, measure the antenna and rear-axle lever arms, complete online mounting-angle calibration, verify wheel-speed scale and time synchronization, and then conduct outage tests in underground garages, tunnels, or building-obstructed routes. Only then can the suitability of an integrated navigation solution for the target vehicle be properly assessed.
📘 Recommended Reading
How Multi-Frequency GNSS Eliminates Ionospheric Delay
Ionospheric delay is one of the major error sources in GNSS positioning. Learn how multi-frequency GNSS measurements help reduce its impact and improve positioning accuracy.
Continue the Conversation
Have questions about this topic or want to share your experience? Join the Qtalis Community to discuss GNSS technologies, exchange ideas, and connect with other users and developers.
Join the Discussion