Tractor Velocity Estimation Using Gyro-data - 06/01/2025 17:56 EST

• Status: Open
• Prize: €500
• Entries Received: 12

Contest Brief

We are looking for a skilled data scientist or machine learning expert to develop a model capable of accurately estimating the velocity of a tractor using gyroscope sensor data. The model must demonstrate stability over long periods and adaptability to diverse road conditions, including straight paths, mountainous terrain, and various surface types (soft, hard, etc.).

Participants must demonstrate their model’s performance using the provided test datasets before being asked to submit detailed code and documentation.

Project Goals
- Develop a robust and accurate model to estimate or compute vehicle velocity from gyroscope data.
- Ensure the model performs reliably across long durations and diverse road conditions.
- Demonstrate the model’s capability by estimating or computing velocities for the provided test dataset.
- Ensure the error range of the model’s velocity estimates is within ±0.5 km/h.

Data Details
1. Gyroscope Sensor Data:
- The dataset includes the following signals:
- Acceleration: AX, AY, AZ
- Angular velocity: Pitch Rate, Roll Rate, Yaw Rate
- The acceleration signals (AX, AY, AZ) are likely scaled by a factor of 10, where 1 g is represented as approximately 9800.
- Angular velocity signals (Pitch Rate, Roll Rate, Yaw Rate) are likely scaled by 10,000.
- Note: The exact scaling factors may require verification and adjustment during model development.

2. Velocity Data:
- Recorded in kilometers per hour (km/h).

3. Time Data:
- Non-uniform sampling at ~50 Hz.
- Each timestamp is recorded in the Time column in milliseconds.

4. Datasets Provided:
- Training Input Data: Includes 32 files, where each file contains time, AX, AY, AZ, Pitch Rate, Roll Rate, and Yaw Rate.
- Training Labels: Includes 32 files, where each file contains the corresponding vehicle velocities (Speed) for the same file in the training input data.
- Test Input Data: Includes 2 files, where each file contains time and gyroscope signals (AX, AY, AZ, Pitch Rate, Roll Rate, Yaw Rate) similar to the training input data.

Participation Process
1. Initial Submission:
- In your initial submission, you need to provide:
- An image of your model’s performance (vehicle speed estimation in km/h) on the provided test input files.
- A brief description of your approach and why it is suitable for this task.
- Only this information is required in the initial submission. Models with promising results will be contacted for further details.

2. Follow-up Submission (for selected participants):
- If the initial submission is validated, you will be asked to provide:
- Model and Code:
- The model’s code and details about the development process.
- Clearly document any pre-processing, scaling, or transformation steps.
- A function that takes gyroscope data as input and returns the speed in km/h.
- Detailed Documentation:
- A description of the methodology and any additional insights into the data or approach.

3. Final Evaluation:
- The submitted model will be tested to ensure:
- Predictions fall within an error range of ±0.5 km/h.
- Stability across long durations and diverse road conditions.
- If the model meets all criteria, you will win the contest.

Evaluation Criteria
- Accuracy: How well the model predicts velocity compared to the ground truth (for the test set).
- Error Margin: Predictions must fall within an error range of ±0.5 km/h.
- Robustness: The model’s ability to handle diverse road conditions and non-uniform time sampling.
- Stability: Consistent performance over long durations.
- Documentation: Clarity and completeness of the provided model and approach (for selected participants).

Additional Notes
- Creativity and innovative solutions are highly encouraged.
- Participants may adjust or refine the gyroscope scaling factors based on their analysis.
- Any programming language is acceptable; however, C and Python are preferred.

Good luck, and we look forward to your innovative solutions!

Recommended Skills

• C Programming
• Machine Learning (ML)
• Python
• Signal Processing
• Tensorflow