SidewalkPilot-v2.4

SidewalkPilot-v2.4

SidewalkPilot-v2.4 is a PyTorch steering regression model for a small autonomous RC car. It predicts a steering servo angle from 0 to 180 degrees from a single OpenCV BGR camera frame.

This checkpoint belongs to Series 2 raw-BGR. It keeps the v2.1+ raw-BGR inference path and was trained after the D0510 field-failure images from the v2.3 tests were merged into the Series 1/2 dataset.

Model Details

• Developed by: Ram Shreyas Naik Sabavat
• Model type: CNN steering regression model
• Library: PyTorch
• License: Apache 2.0
• Checkpoint: SidewalkPilot-v2.4.pth
• Checkpoint created: 2026-05-19 11:28 PM America/Los_Angeles
• Input: Full OpenCV BGR camera frame
• Preprocessing: BGR -> resize 200x66 -> normalize
• Output: Steering servo angle from 0 to 180
• Series: 2.x
• Output scale: approximately 5..175

Specific Improvements

• Trained after merging the D0510 v2.3 field-run images into the labeled dataset.
• Reduced the D0510 dataset MAE from 4.419 on v2.3 to 3.153 on v2.4.
• Reduced the current evaluation max absolute error from 74.412 on v2.3 to 35.268 on v2.4.
• Kept the raw-BGR inference path and did not return to the legacy v2.0 HSV/CLAHE runtime path.
• Became the second-best Series 2 checkpoint on the current 2,224-label evaluation set, behind v2.4b.

Specific Issues Observed / Remaining

• v2.4 improves the D0510 failure set but does not make Series 2 fully reliable in field driving.
• It remains a steering-only behavioral-cloning model, so it does not predict throttle or confidence.
• It still depends on dataset coverage for lighting, driveway transitions, grass edges, shadows, curved curbs, and road-edge position.
• The model does not understand obstacles by itself; LiDAR or another safety layer must remain higher priority.

Output Meaning

Evaluation Setup

• Eval set: 2,224 labeled images
• Failed samples: 0
• Input format: 200x66, OpenCV BGR
• Output scale: servo angle 0..180
• Error unit: servo degrees
• Score formula: max(0, 100 * (1 - absolute_error / 180))
• Dataset note: evaluation includes D0510 v2.3 field-run images merged after the v2.3 cards were written.

Version Update Categories

Evaluation Summary

Negative signed error means the model is left-biased on average.

Prediction Distribution

Ranking

Field Case Comparison

Current Version Snapshot

• Model: 2.4
• Checkpoint: SidewalkPilot-v2.4.pth
• Checkpoint created: 2026-05-19 11:28 PM America/Los_Angeles
• Full score: 98.176%
• MAE: 3.284 servo degrees
• Median AE: 2.457 servo degrees
• Rank in this card: 1 of 9 listed checkpoints

Intended Use

This model is intended for:

• RC car autonomy experiments
• Sidewalk/path steering research
• Raspberry Pi robotics projects
• Small-scale computer vision control systems
• Testing direct image-to-servo steering regression

Out-of-Scope Use

This model is not intended for:

• Real cars
• Public road vehicles
• Human transportation
• Safety-critical systems
• Fully autonomous deployment without external safety layers

System Context

camera frame
-> resize/normalize image
-> PyTorch steering model
-> predicted servo angle
-> runtime decision logic
-> LiDAR safety override when triggered
-> final steering/throttle/brake command
-> servo + motor controller

LiDAR runs as a higher-priority safety layer:

LiDAR clear -> use model steering
LiDAR obstacle -> LiDAR override mode
LiDAR blocked/too close -> hard brake

Training Data

The model was trained on camera images collected from the RC car driving in sidewalk-like environments. Labels represent steering servo angles from 0 to 180 degrees.

v2.4 uses the Series 1/2 dataset after the D0510 v2.3 field-run images were merged. Those D0510 images came from field failures around turns, right-side road-edge driving, and driveway transitions.

Preprocessing

During inference/evaluation, the Series 2 raw-BGR pipeline is:

camera frame in OpenCV BGR
-> resize to 200x66
-> normalize with (x / 255 - 0.5) / 0.5
-> PyTorch steering model
-> servo angle

Do not use the v2.0 HSV/CLAHE preprocessing path with this checkpoint. The model expects the same raw-BGR tensor path used by the runtime for v2.1 and newer.

Limitations

SidewalkPilot-v2.4 can fail when lighting, sidewalk shape, camera angle, shadows, driveway cuts, curved curbs, grass edges, road-edge position, or evening conditions differ from the training data.

The model predicts steering only. It does not predict throttle, reverse, braking, confidence, obstacle presence, or whether the frame is out-of-distribution.

Safety Recommendation

Do not use this model alone to control a robot. In the original project, LiDAR has priority over the model and can override steering or trigger hard braking.

Model Card Contact

Ram Shreyas Naik Sabavat

Links

• YouTube
• Docs
• Hugging Face
• GitHub
• Parts List
• Twitter
• [Grafana Labs](Adding Later...)
• [Weights and Biases](Adding Later...)
• Email