MPPI-Generic: A CUDA Library for Stochastic Trajectory Optimization

MPPI-Generic is a reference implementation paper without measurable adoption (0 stars, 0 forks, no velocity). While the library provides a useful GPU-accelerated abstraction layer for MPPI variants with pluggable cost functions and dynamics models, MPPI itself is well-established (dating to 2015+), and the primary contribution is engineering (CUDA implementation + API design) rather than algorithmic breakthrough. The novelty lies in the generic abstraction and GPU implementation of known algorithms, not in fundamentally new control theory. Defensibility is low because: (1) no user base or community lock-in, (2) the core algorithms (MPPI, Tube-MPPI, Robust-MPPI) are published and replicable, (3) similar functionality could be readily implemented by frontier labs as part of robotics/control platforms. Frontier risk is high: OpenAI, Google, and Anthropic are increasingly investing in embodied AI and robotics; a generic MPPI CUDA library is a natural component they would build in-house or integrate as a standard subroutine rather than depend on an external unmaintained package. The lack of adoption signals the project has not achieved product-market fit as a reusable component. Without evidence of real-world use, community contributions, or ecosystem effects, this remains a research artifact rather than infrastructure.