Privately Owned Vehicle Work Group Meeting - 2025/01/06 - Slot 1

[m-zain-khawaja]

Agenda

• Sharing of goals of Q1 2025 and update on overall progress
• Discussion of overall execution plan of Q1 2025
• Update on SuperDepth Training
• Update on PathDet dataset curation
• Update on LaneDet dataset curation
• Bugfix on SceneSeg visualization

Discussion

Goals and Plan for Q1 2025

@m-zain-khawaja began by sharing the overall goal to complete the AutoSeg Vision Foundation Model by the end of Q1 2025. The planned development goals were to:

• Complete SuperDepth training by end of January 2025
• Complete PathDet dataset curation by end of January 2025 and PathDet training by end of February 2025
• Complete LaneDet dataset curation be end of February 2025 and LaneDet training by end of March 2025

@m-zain-khawaja also spoke about the development philosophy regarding the AutoSeg Vision Foundation Model and the importance of having multiple independent perception mechanisms to achieve the same world-model outputs to have true redundancy and achieve a MTBF which is safe and suitable for fully autonomous driving.

We are on track to achieve these goals.

SuperDepth Network Training Update

Considerable progress has been made on the main scripts and classes which will be used for training the SuperDepth network, updates and new developments shared by @m-zain-khawaja include:

• enabling data augmentation in augmentations.py to account for depth data
• updating benchmark.py to account for SuperDepth benchmarking
• new data loader for SuperDepth network created in load_data_super_depth.py
• creating a separate class for upstream pre-trained model components which will be used in SuperDepth in super_depth_upstream.py and updating the other parts of the overall network architecture to account for this
• creating a new training class for SuperDepth called super_depth_trainer.py which will be helpful in training the model
• creating the main training loop script for SuperDepth called train_super_depth.py

Remaining tasks include fine-tuning of hyperparameters of the model (e.g. batch size, learning rate, loss function, scheduling, minor model tweaks), and once training is completed, the benchmarking, inference and visualization scripts must be implemented.

PathDet Dataset Curation Update

• @TranHuuNhatHuy is nearly finished with dataset curation of CurveLane dataset and expects to submit a pull request soon
• @siddas27 has submitted his pull request for parsing and processing of the Comma2k19 dataset and is performing minor edits based on feedback related to (inclusion of co-linear points in the drivable path and a reduction of the horizon distance over which the drivable path is visualized so that non-line-of-sight drivable path points are not included)

Dataset curation tracking

• TuSimple - 6,394 training data samples
• CULane - 15,800 training data samples
• CurveLane - (parsing completed by @TranHuuNhatHuy, data to be uploaded soon)
• BDD100K - (parsing in progress by @sarun-hub)
• ROADWorks - (parsing in progress by @docjag)
• Comma2k19 - (parsing completed by @siddas27, data to be uploaded soon)

Literature Review

@TranHuuNhatHuy also shared an update on a Literature Review which he has performed on different methods which could be used in the PathDet network to predict the drivable path.

LaneDet Dataset Curation Update

• @devang-marvania has completed the dataset curation and parsing for the LaneDet network using the TuSimple dataset and has submitted his PR and is currently uploading data.

Dataset curation tracking

• TuSimple - 6,394 training data samples
• CULane - to be started
• CurveLane - to be started

Visualization Bug Fix

Alex fixed a bug in the visualization.py file - PR link

Attendees

• @m-zain-khawaja (Senior Tech Lead at AWF and Privately Owned Vehicle Work Group Chair)
• @mitsudome-r (Chair of Autoware Technical Steering Committee)
• @TranHuuNhatHuy (Masters Robotics Student at Tokyo Institute of Technology)
• @siddas27 (ADAS Engineer at Canoo)
• @devang-marvania (Lead Engineer - User Experience, Calibration at Gentherm)
• Alexander Korvyakov (Senior Product Manager - MOIA by Volkswagen)

Zoom Meeting Video Recording

Video Meeting Link
Please contact the work group lead (@m-zain-khawaja) to request access to a recording of this meeting.