Fall 2024 Tasks!#

This is the To-Do list for Fall 2024, outlining both must-have and nice-to-have tasks for the GRiD project.

Must Haves#

[ ] Get everything working on all base cases (Expected 10/25)
[ ] Integrate all code from the academic year into mainline A2R/GRiD
[ ] Continue adding documentation for each function (Google Test, Pybind)
[ ] Add Github badges for tests passing and documentation coverage (See example: https://github.com/rl-tools/rl-tools)
[ ] Set up standard testing suite (may not be straightforward)
[ ] Test CUDA vs. Python automated testing to confirm correct values
[ ] Add Python wrappers for calling CUDA
[ ] GPU autointegration with testing? Docker?
[ ] Ensure Python working bindings in the bindings directory

[ ] 2nd Order Dynamics Gradients (IDSVA) in Python for floating base and CUDA on the GPU
[ ] Complete ABA, CRBA implementations on GPU
[ ] Test and provide documentation for the usage of external forces
[ ] Merge PR for RBDReference algorithms
[ ] Test algorithms on Atlas and fix memory issues (if any)
[ ] Integrate with Trajopt
[ ] DDP GPU vs. CPU comparison:
[ ] Address memory issues and debug mode
[ ] Handle memory challenges for large URDFs (Does Atlas work with GRiD codegen?)
[ ] Optimize memory (e.g., F matrix in minv)
[ ] Build automated testing for GRiD with Docker and CI

[ ] Begin creating GitHub issues and a massive to-do list (ask team members to add anything)
[ ] Software stack for quadruped with demo
[ ] Stabilize baseline process
[ ] Automate building, testing, and documenting for GRiD
[ ] Follow Harvard contribution guidelines and add workflow to auto-add acknowledgements to people who commit/PR to repo.

[ ] Consider extending work into a paper for conferences (e.g., 2nd order DDP | Parallel DDP paper)
[ ] Potential paper on supporting contact

[ ] Develop a framework for writing portable, efficient CUDA code
[ ] Interactive Jupyter notebooks for demonstrations of code functionality with Python bindings

[ ] Build a fully self-contained Docker environment for reproducible builds
[ ] Add automated benchmarking and profiling pipeline for CUDA and Python performance
[ ] Integrate profiling tools like NVIDIA Nsight or nvprof for CUDA kernel performance

[ ] Launch GRiD according to open-source guidelines with unit testing and benchmarking
[ ] Full online documentation for algorithms, including interactive demos
[ ] Create optimization pipeline and framework for generating efficient, scalable code for GPUs
[ ] Get well-documented testing demos on real hardware (videos and interactive notebooks promoting GRiD/A2R lab packages)

By October 11th

By November 15th

[ ] Complete optimization pipeline for code generation, memory management, and kernel fusion
[ ] Plan for full GRiD launch
[ ] Coordinate paper drafts and potential conference submissions
[ ] Collaborate with the GRiD team to get documentation up and running (Sphinx or text files)
[ ] Begin setting up open source contribution guidelines
[ ] Finish setting up automated testing suite (Python vs. CUDA for all URDFs)
[ ] Implement automatic testing for Python vs. CUDA with Pass/Fail status
[ ] Add GitHub badge for testing: “Your commit has been tested successfully!”
[ ] Finish floating base ABA, CRBA implementations on GPU and have everything tested
[ ] Merge all GRiD forks/branches into the main branch

By November 29th

[ ] Complete optimization pipeline with auto-tuning, memory management, and dynamic scheduling
[ ] Have Sphinx/documentation up-to-date with unit testing framework (Catch2 or Google Test)
[ ] Begin testing on real hardware (quadruped software stack)
[ ] Finalize GRiD’s open-source code guidelines and operational status