bajo = low (... level) = batch mojo

Bajo is a work-in-progress, pure-Mojo simulation engine designed for Reinforcement Learning (RL) environments, spatial computing, and physics simulations. The end goal is to run thousands of environments concurrently with high throughput on a single GPU, similar to frameworks like NVIDIA Warp and the Madrona Engine.

Because Mojo is still relatively new, I am currently building out the foundational GPU primitives required for physics simulations from scratch.

Common commands:

pixi run test              # run all tests
pixi run bench_all         # run all benchmarks

Examples:

pixi run example_lbvh

Runs the GPU LBVH normal-rendering example. It should produce the following image

for a detailed version see roadmap

1. Math primitives (Vec, Mat, Quat, Ray, Hit, missing Spatial)
2. obj parser (single threaded done)
3. GPU sort (Bitonic, Radix, Onesweep, but only uints, missing segmented)
4. CPU BVH (Bounds, Triangle BLAS, Sphere BLAS, TLAS)
5. GPU BVH (Bounds, Triangle BLAS, Sphere BLAS, TLAS)
6. GPU Hash Grid (not started)
7. Particle Simulation (not started)
8. SPH Fluid Simulation (not started)
9. Mesh / Particle Coupling (not started)
10. Rigid Body Simulation (not started)
11. Batched Simulation Environments (not started)

see sources

1. struct gpu buffers, not just dtype buffers
2. Parametric Traits (to simplyfy typed bvh)
3. enums