⚡ electrical engineering

A Comparison of Road Grade Preview Signals from Lidar and Maps

This paper demonstrates that a real-time road grade estimation system using on-board lidar and Kalman filtering achieves precision comparable to GNSS/INS-based maps, offering autonomous vehicles a robust, redundant alternative for proactive control when map data is unavailable or inaccurate.

Original authors: Logan Schexnaydre, Aman Poovalappil, Darrell Robinette, Jeremy Bos

Published 2026-04-13

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Logan Schexnaydre, Aman Poovalappil, Darrell Robinette, Jeremy Bos

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are driving a car up a steep hill. If you only realize the hill is there once your front tires hit it, your car has to react instantly—slowing down, shifting gears, or stiffening the suspension. It's a bit like trying to catch a ball you didn't see coming; you might drop it or fumble the catch.

This paper is about giving the car "superpowers" to see the hill before it gets there, so it can prepare in advance.

The Problem: Two Flawed Ways to "See" the Road

Currently, self-driving cars try to guess the road's slope (grade) in one of two ways, and both have problems:

The "Blind" Approach: The car waits until it's actually driving over the hill to feel it. This is like driving with your eyes closed and only knowing you're going uphill when your engine starts to struggle. It's too late to prepare.
The "Map" Approach: The car looks at a pre-made digital map that says, "There is a hill coming up." This is better, but maps can be outdated, missing, or just plain wrong. It's like following a map from 1990 to find a new coffee shop that opened last week; the map might send you to the wrong place.

The Solution: The Car's "Eyes" (Lidar)

The researchers at Michigan Tech wanted to see if a car could use its own Lidar sensors (lasers that scan the environment like a bat uses sonar) to "see" the road ahead in real-time.

Think of the Lidar as a high-tech flashlight that shoots thousands of invisible laser beams. These beams bounce off the ground and return to the car, creating a 3D point cloud—a digital sketch of the road's surface.

How They Made It Work: The "Wheel" Trick

To figure out the slope, the car doesn't need to scan the whole mountain. It just needs to know the height difference between its front wheels and its rear wheels at any given moment.

The Box Filter: Imagine drawing two invisible boxes on the ground: one right where the front tires will touch, and one where the rear tires will touch.
The Laser Scan: As the car drives, the Lidar scans these boxes. It asks, "How high is the ground in the front box? How high is the ground in the back box?"
The Calculation: If the front box is higher than the back box, the car is going uphill. If the back is higher, it's going downhill.

The Glitch: The "Time Lag"

Here is the tricky part. The car moves while it's scanning. Sometimes the laser hits the front tire's spot first, and a split second later, it hits the rear tire's spot. Because the car moved in that split second, the math gets a little messy. It's like trying to measure the length of a moving train by measuring the front and back at different times; you might get the wrong answer.

The researchers solved this with two tools:

A "Time-Travel" Correction: They created a formula to guess how much the car moved between the two measurements and fixed the math.
The "Smart Filter" (Kalman Filter): Think of this as a very smart editor. The raw data from the lasers is a bit noisy (like a radio with static). The Kalman Filter smooths out the static, predicting the most likely true slope based on the car's movement history. It filters out the "jitter" caused by the car's wheels bouncing or the GPS being slightly off.

The Big Test: Lidar vs. The Map

To see if this new "Lidar Eye" worked, they drove a Chrysler Pacifica around a test track with hills, overpasses, and bumps.

Team A (The Map): Used a super-precise GPS and gyro system to create a "gold standard" map of the road.
Team B (The Lidar): Used the new laser system to guess the road shape in real-time.

The Result:
The Lidar system was incredibly accurate. Its guesses were almost identical to the "gold standard" map.

Accuracy: On average, the Lidar was off by only 0.6 degrees. To put that in perspective, that's less than the tilt of a slightly slanted picture frame.
Range: The system could "see" and measure the road about 52 meters (170 feet) ahead. At highway speeds, that gives the car about 3.5 seconds to prepare for a hill.

Why This Matters

This is a game-changer for self-driving cars because:

Redundancy: If the map is missing or wrong, the car doesn't have to guess. It can use its own eyes.
Proactive Driving: Instead of reacting to a hill, the car can anticipate it. It can shift gears early to save gas, soften the suspension to keep passengers comfortable, or brake gently to stay safe.
Independence: It makes the car less dependent on perfect maps, which is crucial for driving in new cities or areas where maps haven't been updated.

The Bottom Line

The researchers proved that a car can use its own laser sensors to "feel" the road ahead just as well as a perfect map can. It's like giving the car a pair of eyes that can see the future slope of the road, allowing it to drive smoother, safer, and more efficiently, even when the map is wrong or missing.

1. Problem Statement

Road grade (slope) is a critical variable for optimizing the safety, energy efficiency, and ride comfort of autonomous vehicles (AVs) and automated systems (e.g., adaptive cruise control, active suspension, and hybrid energy management). Current control systems rely on one of two assumptions regarding grade data, both of which have limitations:

Proprioceptive (Reactive): The vehicle only knows the grade after traversing the segment using onboard sensors (GNSS/INS). This prevents proactive compensation.
Map-Based (Pre-knowledge): The vehicle relies on pre-made Digital Elevation Maps (DEMs). While effective, maps can be outdated, unavailable for new regions, or imprecise, leading to suboptimal control performance.

The paper addresses the need for a real-time, on-board preview signal of road grade that does not rely on external maps, thereby increasing system redundancy and independence.

2. Methodology

The authors propose a Lidar-based Grade Estimation System that estimates road grade in real-time by measuring the elevation difference between the front and rear wheel contact patches.

Core Algorithm

Geometric Definition:
- The instantaneous grade $\theta$ at a waypoint is calculated as $\theta = \sin^{-1}(z/w)$ , where $z$ is the height difference between front and rear wheels, and $w$ is the wheelbase.
- The system defines 3D rectangular "contact patches" for the front and rear wheels based on the vehicle's pose (position and yaw).
Point Cloud Processing:
- Lidar point returns are transformed from the lidar frame to the world frame using GNSS/INS data ( $T^L_W$ ).
- Points falling within the front and rear contact patch boxes are accumulated.
- Once both patches have at least one point, the height difference is estimated by averaging the $z$ -coordinates of points in each patch.
Bias Correction:
- Odometry Bias: Due to the time delay between the front and rear patches acquiring their first measurements, odometry errors accumulate.
- A piecewise linear bias correction function, $b(\Delta f_i)$ , is applied based on the number of frames ( $\Delta f_i$ ) elapsed between the first measurements of the two patches. This is determined empirically.
Kalman Filtering:
- To mitigate the effects of odometry uncertainty and lidar noise, a Kalman Filter is applied in a spatially-indexed domain.
- State Vector: $x = [\theta, \dot{\theta}]^T$ (grade and grade rate).
- Model: A constant grade velocity model is used.
- Noise Modeling: The measurement noise ( $\sigma_r^2$ ) combines lidar range uncertainty and odometry uncertainty. The process noise accounts for the variance in the rate of change of the grade.
Computational Complexity:
- The algorithm's complexity is linear, $O(N \cdot \max(1, d/\Delta s))$ , where $N$ is the number of points, $d$ is the max preview range, and $\Delta s$ is the waypoint spacing. This makes it suitable for real-time implementation.

3. Experimental Setup

Vehicle: Chrysler Pacifica equipped with a Velodyne VLP-32 Lidar and a NovAtel PwrPak7 GNSS/INS receiver.
Location: American Center for Mobility Highway Loop (includes overpasses, underpasses, and hilly terrain).
Reference Ground Truth: A map-based grade signal was generated by recording the pitch output of the GNSS/INS unit at 1-meter intervals.
Parameters:
- Max preview range ( $d$ ): 75m.
- Wheelbase ( $w$ ): 3.09m.
- Contact patch size: 0.5m (to ensure sufficient point returns despite sparse scan patterns).
- Vehicle speed: ~15 m/s.

4. Key Results

The lidar-based estimator was compared against the INS-generated map:

Accuracy & Precision: The lidar-based system produced an unbiased error with a mean of -0.01° and a standard deviation of 0.60°.
Comparison to Maps: The performance is statistically similar to the map-based system. The error distribution fits a Gaussian curve, indicating the lidar system is a valid substitute when maps are unavailable.
Preview Range: The system generated estimates with an average effective range of 52.7 meters. At 15 m/s, this provides a 3.5-second preview for the control system.
Filter Performance: The Kalman filter residuals were zero-mean with small autocorrelation (white noise), confirming the filter accurately models the system dynamics.
Execution Time: The algorithm ran on an AMD Ryzen Threadripper 3960x with execution times ranging from 7.6 ms to 112.9 ms, averaging 58.6 ms. The variance is due to the increasing number of waypoints processed as the vehicle moves forward.

5. Contributions and Significance

Validation of Lidar for Grade Preview: The study demonstrates that automotive lidar can provide road grade previews with precision comparable to high-end inertial map systems, without requiring pre-mapped data.
Enhanced Redundancy: By providing an independent source of grade information, AVs can maintain high-performance control even when maps are missing, outdated, or inaccurate.
Proactive Control Enablement: The system enables proactive control strategies (e.g., gear shifting, suspension adjustment, energy optimization) by providing a 3.5-second lookahead window.
Hybrid Control Strategy: The authors suggest a fusion approach where maps provide a coarse global plan, and the lidar system provides fine-grained local adjustments, or acts as a standalone solution in unmapped areas.
Bias Mitigation Technique: The paper introduces a specific, empirically derived method to correct for odometry-induced bias caused by the temporal delay in acquiring front/rear wheel data, significantly improving estimation accuracy.

Conclusion

The paper concludes that lidar-based grade estimation is a robust, real-time alternative to map-based systems. It successfully mitigates sensor noise and odometry errors through a combination of geometric bias correction and Kalman filtering, offering autonomous vehicles a critical layer of redundancy and independence for grade-sensitive control tasks.