Major improvements to ride trimming, leading to improved tracking. You can expect:

• 7.2% improvement in coaster shape.

• 1.1% improvement in max height MAE.

• 2.0% improvement in max speed MAE.

• 5.8% improvement to track length estimation.

• 2.6% improvement to inversion count.

• 1% improvement to the launch search algorithm.

Added a new mechanism to detect the situation where the user left the phone in the locker but was riding with the watch. The algorithm will now attempt best-effort switching to the watch-based sensors. Note that, generally, watch recordings generate lower quality tracks.

Improved handling for some phone models that do not directly provide altitude estimation from the barometer.

Improved handling of recordings with large gaps in IMU data.

Reduce size of analysis for faster downloading without affecting quality of results.

Improved handling when uncalibrated data from Android is unavailable.

Coaster analyses are now 14% faster on average thanks to some low level optimisation in the algorithm to vectorise some operations

Improved user-facing error and warning messages with explicit reasons for potential analysis degradation along with actionable suggestions.

Max G-force accuracy improved by 34%, reducing mean error from −0.58g to −0.13g against literature values.

Fixed another critical bug where standalone recordings from the Apple watch do not get processed properly.

Fixed an issue where Android recordings with uneven IMU measurements do not get processed properly and resulted in an error 51.

Supported coaster duration has been increased from 8 to 12 minutes. Note that this duration refers to the detected roller coaster segment. The raw recording duration can be much longer.

Improved trajectory optimisation by allowing an additional centre of gravity offset as a degree of freedom.

Improved orientation estimates with more intelligent, dynamic AHRS hyper-parameter selection based on ride regime. This has been enabled by a new open source initiative I have been pushing forward as part of Sensor Logger. See https://sensorlogger.app/zoo.

With these algorithmic changes, you can expect the following improvements:

• For recordings made using Android:

  • 4% improvements in shape quality.

  • 11.5% improvement in coaster length improvement.

  • 3% improvement in max speed estimations.

  • 2% improvement in max height estimations.

  • Negligible change in max g-force estimations.

• For recordings made using iPhones:

  • 0.6% improvement in shape quality.

  • 1.3% improvement in max height estimations.

  • 1% improvement in coaster length estimations.

  • 0.7% improvement in max speed estimations.

  • Negligible change in g-force and inversion estimations.

• Generally, for tracks in GPS-denied environments on either platform, you can specifically see on average:

  • 3.4% improvement in shape quality.

  • 6.5% improvement in coaster length estimations.

  • 2.7% improvement in max height estimations.

  • 1% improvement in max speed estimations.

  • 2.2% improvement in g-force estimations.

Fixed a critical bug where standalone recordings from the Apple watch do not get processed properly.

Introduce Android support! Coaster Logger and Coaster Fusion now support Android! Note that quality of tracks from Android phones may vary in quality more than the iPhone counter parts — this is because of the heterogeneous nature of the Android ecosystem. Further, your Android phone must support the minimum set of sensor requirements — Accelerometer, gyroscope, barometer and GPS. Microphone is optional.

The Attitude and Heading Reference System (AHRS) algorithm has been completely redeveloped, providing better performance consistency across heterogenous smartphone hardware. The custom sensor fusion algorithm handles magnetometer signals more intelligently, especially under challenging magnetic environments, such as launches and break segments. The custom filter also generally behaves in a more stable way under high dynamic conditions.

With the new AHRS changes, you can expect a 5% average improvement in track shape — but most importantly, the worst case scenario can improve by up to 50%. Height and speed estimations also improve in accuracy by 3.6% and 4.7% respectively. Launch and lifthill counts also improved marginally. These values are applicable to tracks measured with iPhones.

Improved handling and warning when there are unexpected gaps in non-inertial data such as acceleration or gyroscope measurements.

Improved coaster trimming heuristics when barometer values are unreliable.

Improved max acceleration and max speed estimations.

Minor performance improvements.

Improved reward function when integrating acceleration information, taking into account barometer information:

• iPhone enhancements, specifically in GPS-denied environments (e.g. indoor rides) (415bfb18)

  • 6% improvements in coaster shape estimation.

  • 4.5% improvements in speed estimation.

  • 7.6% improvements in coaster length estimation.

• Apple Watch specific enhancements (37f150f1)

  • A 42% reduction in the RMSE of maximum height estimation.

• Overall improvement across the entire validation dataset (d8b4305c)

  • General coaster shape expected to improve on average by 3%, with minimal improvements to height estimation (<1%).

  • 3.5% improvement in speed estimation expected.

More tolerant to large and variable gaps in barometer measurements, especially from Apple Watches. However, note that the altitude information will be degraded accordingly.

Enhanced walking detection algorithm to improve recording trimming, allowing the analysis the focus on the true roller coaster portions.

Fixes a critical issue where Watch Only recordings of coasters will not parse at all.

Improved representative point calculation, using all available GPS points during the recording, not just the ride portion.

Fixes a critical issue with Watch recordings exhibiting incorrect altitude fusion due to coordinate system mismatch.

Initial release.