At PQS Tech, our approach to the hardware we supply and support is rooted in verification. For a professional surveyor, a point cloud is only as reliable as the methodology used to generate it.
To evaluate the current state of mobile mapping, we conducted a comparative study in our office, benchmarking several SLAM systems, including the CHCNAV RS30, CHCNAV RS7, Viametris, and SHARE S20, against our trusty Trimble X7 terrestrial laser scanner (TLS).
The Key Takeaway
Our study confirms that when SLAM data is treated with traditional survey discipline - utilising Trajectory Ground Control Points (TGCPs) and a stabilised gait - it can achieve a Mean Error in Plane of 5mm, rivalling TLS for many BIM and interior mapping applications. Without these constraints, rotational drift can render the data unusable for professional tolerances, even though it may ‘look’ great.
The Survey Methodology: Establishing the "Truth"
To ensure a rigorous comparison, we did not rely on "best-fit" cloud-to-cloud registration. We followed a traditional survey workflow to isolate hardware performance from processing artefacts:
-
Primary Control: We established a rigid control network using a Total Station to define the absolute coordinate system, transferring a baseline from outside into our office upstairs.
-
The TLS Benchmark (Green): We captured the environment using a Trimble X7. By performing a resection in the centre of the largest room, we ensured the control points maintained good geometry. This dataset serves as our "Truth", representing static observations with negligible drift.
-
Controlled SLAM Capture (Orange/Red): For the CHCNAV RS30 and SHARE S20, we utilised a controlled trajectory. We integrated Trajectory Ground Control Points (TGCPs) - surveyed coordinates that "pin" the SLAM path to the primary network during post-processing.
-
Uncontrolled SLAM Capture (Purple/Yellow): To simulate poor field practice, we captured the office without TGCPs and without optimised trajectory loops.
-
Post-Processing: Data was processed in proprietary software for each manufacturer, with the final comparison conducted in Trimble RealWorks.
The Fundamental Difference: Static Observations vs Dynamic Draping
The core difference between TLS and SLAM lies in the spatial reference of the sensor. A TLS provides rigid observations from a fixed position. In contrast, SLAM data is effectively sensor data "draped" onto a calculated trajectory.
The SLAM algorithm estimates the sensor’s Pose in real-time using an Inertial Measurement Unit (IMU) and LiDAR feature tracking. However, IMUs naturally drift over time. If the trajectory estimation drifts, even by a fraction of a degree, the error is propagated through the entire point cloud. Without external constraints like TGCPs and a sensible walking path, this leads to the "ghosting" or "doubling" of walls typically seen in uncontrolled datasets.
Trajectory Discipline: Beyond the "Walk-and-Go" Myth
A SLAM system is fundamentally reliant on the quality of the sensor path (the trajectory). For this project, we executed a disciplined, stabilised trajectory with a total collect time of 3 minutes and 12 seconds.
By maintaining a smooth, consistent gait and a deliberate route, we ensured the sensors maintained optimal feature tracking. We didn't treat the scanner like a consumer camera; we treated it as a survey instrument. This professional approach ensures the sensor has sufficient time to "see" geometric anchors—such as corners, door frames, and alcoves—without the jarring motion that causes algorithmic failure.
I explored poor methodology versus the technology in detail in a recent article, found here.
Visual Analysis: Comparative Performance
When analysing the datasets side-by-side, the impact of trajectory anchoring and a disciplined route becomes undeniable.
1. Global Alignment and Planimetric Consistency
In the plan view, the Green (X7) data acts as our primary control.
-
The Orange (RS30 + TGCPs) and Red (SHARE S20) datasets show high planimetric fidelity, following the X7’s wall profiles with minimal deviation.
-
The Purple (Viametris) and Yellow (Uncontrolled) begin to diverge significantly. This is rotational drift, where the entire room geometry begins to skew relative to the true office layout.
2. Wall Thickness and "Smearing"
Looking at a close-up of a partition adjoining the external brickwork, we can analyse the "thickness" of the wall data:
-
Green: A clean, single-skin line (the benchmark).
-
Orange/Red: Maintains a tight overlap, confirming the trajectory was anchored well enough to prevent "smearing."
-
Yellow/Purple: Shows significant smearing where the algorithm has misjudged the true movement of the sensor. This most likely occurred whilst transitioning through a door frame, where the sensor saw limited distinct features.
3. Vertical Profile Fidelity
While relative accuracies within a single room are often maintained in uncontrolled SLAM, absolute accuracy is degraded substantially. As the trajectory drifts, the vertical profile begins to shift or tilt, leading to significant errors in height measurements on larger projects.
Summary of Results
Conclusion: Methodology Over Hardware
The results of our controlled run with the RS30 show that 100% of the points met a precision threshold of 0–1cm, with a Mean Error in Plane of 5mm and a Mean Error in Height of 6mm.
These results are only possible when the data is treated as a trajectory-based observation requiring both external constraints and professional gait.
When to use what?
-
Use TLS when: You require sub-3mm accuracy for structural monitoring, high-precision engineering, or floor flatness (FF/FL) analysis.
-
Use SLAM when: Speed is critical, environments are congested, or you are performing high-detail BIM/as-built surveys - provided you have the expertise to anchor the trajectory.
At PQS Tech, we don’t guess. We verify.
How are you managing trajectory drift in environments with limited vertical features? Do you find that increasing TGCP density compensates for poor loop-closure opportunities, or do you rely more on post-processing optimisation? Let’s discuss the nuances of trajectory anchoring in the comments.
0 comments