Pressure Drop: Gravity Vector Estimation from a Distributed Barometric Pressure Sensor Array

William Dubois, Nicolas Lauzon, Johann Laconte, François PomerleauarXiv 2026 · arXiv

TL;DR

Researchers have developed a novel method to estimate a robot's tilt (the gravity vector) using an array of barometric pressure sensors instead of traditional IMUs. By measuring relative elevation differences across a distributed array, the system provides a direct observation of gravity that remains accurate even during rapid, high-acceleration maneuvers where standard sensors often fail.

What Problem Does This Paper Address?

In mobile robotics, knowing which way is "down" is crucial for stable navigation and accurate mapping (SLAM). Traditionally, we rely on Inertial Measurement Units (IMUs). However, accelerometers in IMUs don't measure gravity directly; they measure "proper acceleration"—a combination of gravity and the robot's own motion.

As shown in Figure 1, when a robot accelerates quickly, turns sharply, or vibrates, the accelerometer signal becomes "polluted" by these motions, making it difficult to isolate the gravity vector. While filters can mitigate this, they often struggle with high-dynamic scenarios or sensor saturation. This paper tackles the challenge of finding a motion-independent way to observe gravity.

Fig. 1: Field robots undergo rapid motions and forces that can challenge the IMU for estimating gravity. For instance, the accelerometer measurement from the IMU, aimu corresponds to the proper acceleration, combining the robot acceleration (arobot) and the gravity (g) reaction force. We propose a novel way to estimate tilt (ϕ), the angle between the z-axis of the robot frame (z) and g, via pressure sensor measurements (pi) around origin O, independent of the current motions of the robot.

Methods

The authors propose using a distributed array of barometric pressure sensors. The core idea is simple but elegant: barometric pressure correlates with elevation. If you place multiple sensors at known positions on a robot, the differences in pressure between them reveal their relative heights. These height differences are directly tied to the robot's orientation relative to the gravity vector.

Key technical components include:

  • The Hypsometric Equation: Used to convert pressure ratios into differential elevation ($\delta z$).
  • Linear Estimation: They formulated a linear system $z = Pg$, where $z$ is the vector of measured elevations, $P$ is the known geometry of the sensor array, and $g$ is the gravity vector to be estimated.
  • Observability Analysis: The authors proved that while four non-coplanar sensors provide a unique solution, as few as three sensors can work if you assume the robot isn't upside down.
  • Noise Modeling: A theoretical evaluation of how sensor placement (intersensor distance and geometry) affects the covariance of the tilt estimate.

Datasets

The study relied on custom-collected experimental data rather than standard benchmarks:

  • The Rig: A custom gravity estimation rig equipped with eight DPS310 barometric sensors, a high-precision industrial reference pressure sensor (Setra 278), and two different grades of IMUs (Xsens MTi-10 and CHR-UM6).
  • Ground Truth: A Vicon Tracker motion capture system provided millimeter-accurate pose data for validation.
  • Field Data: Four hours of data across 13 experiments, including both static and highly dynamic maneuvers (high-G turns and accelerations).
  • Mapping Test: An outdoor deployment on an Uncrewed Ground Vehicle (UGV) to test integration with an Iterative Closest Point (ICP) mapping algorithm.

Key Insights and Ideas

  • Motion Independence: Unlike IMUs, pressure-based tilt estimation is theoretically independent of the robot's linear acceleration and rotational velocity. This makes it a perfect redundant sensor for "noisy" dynamic environments.
  • Geometry Matters: The authors found that placing sensors on a plane orthogonal to gravity (e.g., the roof of a vehicle) significantly outperforms placing them on a vertical plane. This is because it minimizes the covariance in the tilt estimate.
  • The "Sweet Spot" for Distance: Increasing the distance between sensors improves accuracy (reducing the angular error), but only up to a point. In their experiments, accuracy plateaued around 1.25 meters; beyond that, mechanical flex in the robot's frame began to introduce new errors, breaking the "rigid body" assumption.
  • Averaging the Origin: Using the average of two sensors as the "origin" for the array significantly reduced noise, improving performance by up to 28%.

Key Results

  • High Precision: In semi-static scenarios, the system achieved a median tilt error of just 0.87°.
  • Robustness to Dynamics: As shown in Figure 5 (below), while IMU-based filters (like Madgwick or VQF) saw their errors spike as acceleration increased—especially when the low-cost CHR-UM6 saturated—the pressure-based approach maintained consistent performance regardless of the "G-force" applied.
  • SLAM Integration: Qualitatively, the system successfully provided the "initial guess" for an ICP-based mapping algorithm, producing high-quality 3D maps of outdoor environments with significant elevation changes.
Fig. 5: Impact of the acceleration magnitude on the tilt esti- mation angular error in highly dynamic scenarios.

Limitations and Future Work

  • Environmental Sensitivity: Barometric sensors are sensitive to wind currents and weather changes. While the authors used differential measurements to cancel out slow atmospheric shifts, rapid wind gusts (e.g., from the robot's own movement or external wind) could still induce errors.
  • Rigidity Requirements: The method assumes the sensor array is perfectly rigid. Any flexing of the robot's chassis at high speeds or over rough terrain introduces "pseudo-elevation" changes that degrade the estimate.
  • Future Directions: The authors suggest investigating the impact of vehicle speed on pressure readings (Bernoulli effect) and tighter fusion of this pressure-based gravity vector with IMU data for a truly "all-weather" orientation filter.

Summary updated

Abstract

While gravity is often used in mobile robotics as a global constraint for Simultaneous Localization and Mapping (SLAM) algorithms, most estimation methods rely on indirect observation of the gravity vector. In this paper, we propose the use of altimeter measurements to directly observe the direction of the gravity vector. We provide a theoretical evaluation of the estimation noise and validate it through extensive experimental data, along with an evaluation of the impact of intersensor distance on our gravity vector estimates. Experimental results demonstrate that a median error of $0.87^\circ$ on the tilt estimation can be achieved with our system in semi-static scenarios. Furthermore, we demonstrate the potential of this approach by comparing it with state-of-the-art methods based on Inertial Measurement Units (IMUs). We highlight that we achieve similar performance to that of IMU-based systems in highly dynamic scenarios and even outperform in specific cases of accelerometer measurement saturation. Finally, an additional experiment shows the potential use within a robotic system through the integration into an Iterative Closest Point (ICP) algorithm.
BibTeX
@misc{dubois2026pressure,
  author = {William Dubois and Nicolas Lauzon and Johann Laconte and François Pomerleau},
  title = {{Pressure Drop: Gravity Vector Estimation from a Distributed Barometric Pressure Sensor Array}},
  year = {2026},
  eprint = {2611.23120},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.23120},
}
Figure