GNSS for Robotics & Autonomous Delivery

Ground-based autonomous platforms -- delivery robots, agricultural rovers, inspection vehicles, and off-highway machines -- face a positioning problem that is fundamentally different from that of manned vehicles or UAVs. They operate at low speed, traverse mixed indoor-outdoor environments, navigate narrow sidewalks and corridors, and must maintain centimeter-level lane discipline without the benefit of a human operator. This article examines how modern multi-constellation, multi-frequency GNSS receivers combined with inertial navigation systems address the unique demands of outdoor robotics.

The Role of GNSS in Outdoor Robotics

Any robot that operates outdoors needs absolute position in a global reference frame. Odometry drifts. LiDAR-based SLAM accumulates error over long traversals. Camera-based localization fails under changing lighting and weather. GNSS provides the drift-free, absolute position fix that anchors all other sensor modalities to a consistent world frame.

However, standalone single-frequency GNSS is insufficient for robotic applications that demand lane-level or sub-lane-level accuracy. A single-frequency GPS receiver delivers 1.5 to 3 m CEP accuracy in open sky, degrading to 5 m or worse in urban environments. For a sidewalk delivery robot that must stay within a 1.5 m wide path, or an agricultural ground robot that must follow crop rows spaced 76 cm apart, this level of error is unacceptable.

Meeting the positioning requirements of outdoor robotics demands several technology layers beyond basic GPS:

Multi-constellation tracking. Receiving signals from GPS, GLONASS, Galileo, BeiDou, and QZSS maximizes satellite availability, which is critical in urban canyons and near buildings where portions of the sky are obstructed. A receiver tracking 40+ satellites simultaneously is far more resilient than one limited to GPS alone.
Multi-frequency reception. Dual-frequency (L1/L2) or triple-frequency (L1/L2/L5) receivers eliminate ionospheric delay as an error source and enable faster ambiguity resolution for RTK positioning.
RTK corrections. Real-Time Kinematic positioning reduces horizontal error to 1-2 cm RMS in open sky. Corrections can be delivered via a local base station over UHF radio or, more commonly for mobile robots, via an NTRIP caster over the robot's cellular data link.
Inertial aiding. An integrated Inertial Measurement Unit (IMU) provides continuous position and attitude propagation during brief GNSS outages caused by tree canopy, building overhangs, or tunnel passages.

The combination of these technologies -- multi-constellation, multi-frequency GNSS with RTK corrections and INS aiding -- forms the positioning backbone for virtually all serious outdoor robotics applications today.

Delivery Robots & Last-Mile Autonomy

Sidewalk delivery robots represent one of the fastest-growing applications of outdoor autonomous navigation. Companies operating fleets of small wheeled robots for food, grocery, and package delivery face a set of positioning challenges that are distinct from those encountered by autonomous cars or trucks.

A delivery robot typically operates on sidewalks, crosswalks, and pedestrian paths that are 1.2 to 2.5 m wide. The robot itself may be 50 to 70 cm wide, leaving minimal margin for lateral positioning error. To stay safely on the sidewalk and avoid curbs, obstacles, and pedestrians, the robot needs consistent sub-decimeter horizontal accuracy.

RTK-corrected GNSS can deliver this level of performance in open-sky conditions. The challenge is that delivery robots rarely enjoy open sky. They operate in residential neighborhoods lined with trees, on university campuses surrounded by multi-story buildings, and in downtown commercial districts with deep urban canyons.

Urban Canyon Challenges

In dense urban environments, GNSS performance degrades due to several interrelated effects:

Satellite occlusion. Buildings block direct line-of-sight to satellites at low elevation angles, reducing the number of usable signals and degrading geometry (increasing DOP values).
Multipath. Signals reflected off building facades, glass windows, and metal structures arrive at the antenna with additional path delay, corrupting pseudorange and carrier-phase measurements. Multipath is the dominant error source in urban GNSS.
Signal diffraction. Signals that graze building edges are diffracted and arrive with distorted phase, making carrier-phase ambiguity resolution unreliable.

These effects can cause RTK fix availability to drop below 50% in downtown corridors and introduce position errors of several meters even when a fix is reported. For delivery robots, this means GNSS alone cannot be the sole positioning source. It must be fused with complementary sensors -- cameras, LiDAR, wheel odometry, and inertial measurement -- in a robust state estimation framework.

Campus and Suburban Deployment

University campuses and suburban neighborhoods offer a more favorable GNSS environment than dense urban cores. Buildings are typically lower, streets and sidewalks are wider, and tree canopy, while present, is less dense than in forested areas. In these environments, RTK fix availability of 80-95% is achievable with a quality multi-constellation receiver and a well-placed antenna. The remaining gaps are bridged by INS propagation and vision-based localization.

For campus delivery fleets, the positioning system must also handle transitions between outdoor navigation and semi-indoor environments such as building overhangs, covered walkways, and loading docks. A tightly coupled GNSS/INS solution can propagate position through these brief outages without requiring a full re-acquisition sequence.

Autonomous Mobile Robots (AMRs)

The term AMR encompasses a broad class of ground-based autonomous platforms beyond last-mile delivery. Three categories are particularly relevant to GNSS integration.

Warehouse-to-Outdoor Transition Robots

Many logistics operations require AMRs to move goods between indoor warehouse environments and outdoor loading areas, staging yards, or adjacent buildings. Indoors, these robots typically navigate using LiDAR-based SLAM or fixed infrastructure such as magnetic tape or QR codes. Outdoors, they must switch to GNSS-based positioning.

The transition zone -- the loading dock door, the warehouse entrance, the boundary between covered and uncovered areas -- is the most challenging region for the positioning system. GNSS signals are not yet fully acquired, LiDAR maps from the indoor environment no longer apply, and the robot must maintain continuous navigation through the handoff. A GNSS/INS system with fast re-acquisition and robust inertial bridging is essential for smooth transitions.

Agricultural Ground Robots

Small autonomous rovers for crop scouting, targeted spraying, mechanical weeding, and phenotyping operate in structured agricultural environments where row spacing and plant locations are well-defined. These robots demand:

Centimeter-level absolute accuracy to stay within crop rows and avoid damaging plants.
Repeatable pass-to-pass performance so that successive traversals follow the same path, enabling precise application maps and temporal comparisons.
Heading accuracy at near-zero speed for tight headland turns and point-turn maneuvers.

RTK-corrected GNSS with dual-antenna heading provides the positioning performance these applications require. The relatively open-sky environment of most agricultural fields ensures high RTK fix availability, making GNSS the primary position source with minimal reliance on supplementary sensors.

Inspection and Security Robots

Autonomous robots deployed for perimeter security, infrastructure inspection (solar farms, pipelines, substations), and environmental monitoring operate in outdoor environments that may include partial GNSS obstruction from structures, fences, or vegetation. These platforms require reliable georeferenced position for patrol route adherence, anomaly localization, and reporting. A GNSS/INS solution ensures position continuity when the robot passes under covered structures, through tunnels, or alongside large metallic objects that cause multipath.

GNSS/INS Sensor Fusion

The integration of GNSS and inertial navigation is the most important architectural decision in a robotics positioning system. Two primary coupling strategies exist, each with distinct trade-offs.

Loosely Coupled Integration

In a loosely coupled architecture, the GNSS receiver and the INS operate as independent subsystems. The GNSS receiver outputs a position and velocity solution, which the INS Kalman filter consumes as an observation update. This approach is straightforward to implement and allows the GNSS receiver and IMU to be sourced from different vendors.

However, loosely coupled integration has significant limitations for robotics:

It requires a minimum of four satellites for the GNSS receiver to produce a position fix. In degraded environments where only two or three satellites are visible, the GNSS subsystem provides no update at all, and the INS must free-run on inertial data alone.
Multipath-corrupted GNSS solutions are passed directly to the INS filter, which has no visibility into the underlying satellite measurements and cannot reject individual corrupted signals.
RTK re-acquisition after an outage is slower because the INS does not assist the GNSS receiver's ambiguity resolution process.

Tightly Coupled Integration

In a tightly coupled architecture, raw GNSS observables -- pseudorange, carrier phase, and Doppler measurements from each individual satellite -- are fed directly into the INS navigation filter alongside IMU data. This approach offers several advantages critical to robotic applications:

Individual satellite aiding. Even a single visible satellite provides useful information to constrain INS drift. The system does not need a full GNSS fix to benefit from satellite observations.
Measurement-level integrity. The navigation filter can evaluate the consistency of each satellite measurement against the predicted INS solution and reject individual signals corrupted by multipath or diffraction.
Faster RTK re-convergence. The INS-predicted position constrains the search space for carrier-phase ambiguity resolution, enabling faster RTK fix recovery after outages.
Superior bridging performance. Because the system continuously calibrates IMU biases and scale factors using GNSS observables, the quality of the inertial-only solution during outages is substantially better than in a loosely coupled system.

For robotics applications operating in mixed environments -- sidewalks, campuses, agricultural fields with tree lines, industrial facilities -- tightly coupled GNSS/INS is the recommended architecture. The Hemisphere VS-I8 implements tightly coupled GNSS/INS fusion in a compact enclosure suitable for integration on small robotic platforms.

Bridging GNSS Outages with IMU

The duration for which an IMU can maintain useful position accuracy without GNSS updates depends on the grade of the inertial sensors. For the MEMS-grade IMUs typically used in robotics (due to SWaP constraints), useful bridging durations are:

Consumer-grade MEMS (< 10 deg/hr bias stability): 5 to 10 seconds of useful position propagation.
Tactical-grade MEMS (1 to 10 deg/hr): 15 to 30 seconds, sufficient for most tunnel passages and building overhangs.
Navigation-grade MEMS (< 1 deg/hr): 60 seconds or more, adequate for extended covered areas.

For most robotic applications, a tactical-grade MEMS IMU in a tightly coupled architecture provides the best trade-off between bridging capability, size, weight, power consumption, and cost.

Heading at Low Speed

Heading determination is a critical and often underestimated challenge for ground robots. Most GNSS receivers derive heading from the velocity vector -- the direction the receiver is moving. This works well for vehicles traveling above approximately 5 km/h, where the velocity vector is well-defined and aligned with the vehicle's longitudinal axis.

Delivery robots, agricultural rovers, and inspection platforms frequently operate at speeds below 3 km/h, stop frequently, and execute point turns. At these speeds, velocity-derived heading is unreliable or entirely unavailable:

At zero speed, velocity heading is undefined.
At very low speed (< 1 m/s), position noise dominates the velocity estimate, producing heading errors of tens of degrees.
During point turns, the velocity vector is nearly zero while the robot's heading changes by 90 to 180 degrees. The GNSS receiver provides no heading information during the maneuver, precisely when the robot needs it most.

Dual-Antenna Heading

The solution is dual-antenna GNSS heading (also called GNSS compass). Two GNSS antennas are mounted on the robot at a known baseline separation -- typically 30 cm to 1 m, depending on the platform size. The heading is computed from the carrier-phase difference between the two antennas, providing accurate heading regardless of vehicle speed, including at a complete standstill.

Dual-antenna heading accuracy depends on the baseline length and the quality of the carrier-phase measurements. For a 50 cm baseline with multi-frequency RTK, heading accuracy of 0.3 degrees RMS is achievable. For a 1 m baseline, this improves to approximately 0.15 degrees RMS.

For robotics platforms, dual-antenna heading eliminates the fundamental limitation of velocity-derived heading and enables reliable navigation through low-speed maneuvers, point turns, and stationary orientation holds. The Vega 34 compass board provides dual-antenna heading in a compact OEM form factor designed for embedded integration.

Recommended Hemisphere GNSS Hardware

Selecting the right GNSS hardware for a robotic platform requires balancing accuracy, size, weight, power consumption, and cost. The following Hemisphere GNSS products address the range of requirements encountered in outdoor robotics.

Phantom 20/34 OEM Boards

The Phantom 20 and Phantom 34 are compact GNSS OEM receiver boards designed for embedded integration. The Phantom 34 tracks all major constellations on multiple frequencies and supports RTK positioning with centimeter-level accuracy. Its small form factor and low power consumption make it well-suited for battery-powered robotic platforms where SWaP (Size, Weight, and Power) budgets are tight. The Phantom 20 provides a cost-effective option for applications that can operate with fewer constellation and frequency options while maintaining RTK capability.

For robotics engineers designing a custom navigation stack, the Phantom boards provide raw GNSS observables, PVT solutions, and RTK-corrected positions over standard serial interfaces, allowing tight integration with an external IMU and the robot's navigation computer.

VS-I8 GNSS/INS Smart Antenna

The VS-I8 combines a multi-constellation, multi-frequency GNSS receiver, a tactical-grade MEMS IMU, and a tightly coupled GNSS/INS navigation engine in a single ruggedized enclosure. For robotics applications, the VS-I8 offers a turnkey positioning solution that eliminates the need to develop custom sensor fusion software.

Key specifications relevant to robotics include tightly coupled GNSS/INS fusion for robust performance in partially obstructed environments, inertial bridging through GNSS outages, and a compact form factor suitable for mounting on medium-sized robotic platforms. The integrated design reduces cabling, connector count, and potential points of failure compared to assembling separate GNSS and IMU subsystems.

Vega 34 Compass Board

The Vega 34 is a dual-antenna GNSS compass OEM board that provides accurate heading regardless of vehicle speed. For delivery robots and agricultural rovers that operate at walking speed or below, the Vega 34 solves the low-speed heading problem without requiring a magnetometer (which is unreliable near metal structures and electric motors) or a high-grade gyroscope (which is expensive and power-hungry).

The board accepts inputs from two GNSS antennas and outputs heading along with RTK-corrected position. It can be integrated directly into the robot's control system via serial interface.

HA32 UAV Antenna

The HA32 is a lightweight, low-profile multi-frequency GNSS antenna originally designed for UAV applications. Its compact size and light weight make it equally suitable for small ground robots where antenna mounting space is limited. The HA32 provides full multi-constellation, multi-frequency coverage (GPS L1/L2/L5, GLONASS G1/G2/G3, Galileo E1/E5a/E5b/E6, BeiDou B1/B2/B3) in a form factor that does not significantly impact the robot's SWaP budget.

For dual-antenna heading configurations on small platforms, two HA32 antennas can be mounted at a fixed baseline separation to feed a Vega 34 compass board, providing both RTK position and heading in a minimal footprint.

Phantom 40 OEM Board

The Phantom 40 is Hemisphere's higher-performance OEM receiver, designed for applications that demand the best available GNSS accuracy and signal tracking capability. For robotics platforms that operate in particularly challenging GNSS environments -- dense urban areas, heavy tree canopy, or near large reflective structures -- the Phantom 40 provides improved tracking sensitivity, faster RTK convergence, and more robust multipath mitigation compared to the Phantom 20/34.

The Phantom 40 is recommended for robotics applications where positioning accuracy and availability are mission-critical and the SWaP budget can accommodate a slightly larger receiver board. Off-highway autonomous vehicles, large inspection robots, and high-value agricultural platforms are typical use cases where the Phantom 40's enhanced performance justifies the additional investment.

Selecting the Right Configuration

The optimal hardware configuration depends on the specific robotic application:

Application	Recommended Receiver	Heading Solution	Antenna	INS Required
Sidewalk delivery robot	Phantom 34	Vega 34 (dual-antenna)	HA32 (x2)	Yes
Agricultural ground robot	Phantom 34	Vega 34 (dual-antenna)	HA32 (x2)	Optional
Warehouse-outdoor AMR	VS-I8	Internal GNSS/INS	Integrated	Yes
Inspection / security robot	Phantom 40	Vega 34 (dual-antenna)	HA32 (x2)	Yes
Off-highway autonomous vehicle	Phantom 40 + external INS	Dual-antenna via Phantom 40	Full-size survey	Yes

For all configurations, RTK corrections should be delivered via NTRIP over the robot's cellular modem, using a regional CORS network or a commercial correction service. The receiver boards support standard RTCM 3.x input, making them compatible with any standards-compliant correction source.