AI & ML
Laguerre-MPPI
arXiv cs.SY0 views
arXiv:2611.24106v1 Announce Type: new Abstract: Model Predictive Path Integral control (MPPI) is a sampling-based control method that handles nonlinear dynamics and non-convex costs without gradient information. Since MPPI samples the entire input sequence, longer prediction horizons lead to a higher-dimensional sampling space, potentially degrading control performance. This paper proposes Laguerre-MPPI, which parameterizes the control input sequence using discrete-time Laguerre basis functions and samples the basis coefficients. The resulting search dimension is independent of the prediction horizon. Hence, a longer prediction horizon does not require more samples to maintain the sampling density. For linear systems with quadratic costs, we derive a closed-loop stability condition in the infinite-sample limit. Under finite sampling, we provide a closed-form expression for the limit of the ratio of the effective sample size to the number of samples. We also show that the upper bound of this limit for Laguerre-MPPI is no lower than that for conventional MPPI. Numerical experiments demonstrate smoother inputs in obstacle avoidance and maintained performance over long horizons. Moreover, they verify the lower bound on the number of samples given by the finite-sample analysis.
Source:https://arxiv.org/abs/2611.24106PreviousAdaptive Guidance for 3GPP Ambient IoT Contention-Based Random AccessNextKeeping Shielded Data-Driven Controllers Safe Under System Evolution via Model Consistency
Related posts
I Built a Tiny Japanese GPT from Scratch in PyTorch and Tried a Toy 'RSI' Loop: Here's What It Does and Doesn't Show
0dev.to
Which Bots Show Up in Your Analytics? I Sent 10 Automated Visitors to a Test Page
0dev.to
The Smart Ring Goes to Church
0Wired
Muse Is Unleashing a World of Pettiness on Instagram
0Wired

The Last Word: launch article
0dev.to
What people are building, discovered through new domains
0Hacker News