SLAM

Simultaneous Location and Mapping

2014-11-25 — 2022-04-28

Wherein the problem of scene reconstruction by a moving camera is examined, and neural implicit representations with differentiable rendering are combined to enable scalable SLAM via least-squares inference.

algebra
approximation
Bayes
distributed
dynamical systems
Gaussian
generative
graphical models
Hilbert space
linear algebra
machine learning
networks
optimization
probability
signal processing
state space models
statistics
stochastic processes
time series
Figure 1: Estimate of unknown \(\mu\)

Classic robotics problem: reconstruct a scene by moving a camera around the room.

In practice, it often boils down to a least squares inference problem or more generally a Gaussian Belief propagation inference problem.

1 NICE-SLAM

I am interested in a recent technique that combines implicit representation with SLAM (Zhu et al. 2022).

There are many cool tricks combined there — differentiable rendering etc. Hierarchical implicit representations.

2 Tools

2.1 jaxfg

2.2 gradslam

Differentiable GPB solver (Jatavallabhula, Iyer, and Paull 2020).

2.3 ceres solver

ceres-solver, (C++), the Google least squares solver, seems to solve this kind of problem. I am not sure where the covariance matrices go in. I occasionally see mention of “CUDA” in the source repo so maybe it exploits GPUs these days.

2.4 Incoming

3 References

Davison, and Ortiz. 2019. FutureMapping 2: Gaussian Belief Propagation for Spatial AI.” arXiv:1910.14139 [Cs].
Eustice, Singh, and Leonard. 2006. Exactly Sparse Delayed-State Filters for View-Based SLAM.” IEEE Transactions on Robotics.
Jatavallabhula, Iyer, and Paull. 2020. ∇SLAM: Dense SLAM Meets Automatic Differentiation.” In 2020 IEEE International Conference on Robotics and Automation (ICRA).
Zhu, Peng, Larsson, et al. 2022. “NICE-SLAM: Neural Implicit Scalable Encoding for SLAM.”