Mohamad Qadri

I am a PhD student at the Robotics Institute at Carnegie Mellon University, where I work with Michael Kaess in the the Robotics Perception Lab . I work on statistical methods for state estimation. I am interested in developing algorithms for efficient and robust inference at the intersection of linear algebra and probabilistic graphical models.

I completed my Masters in Robotics at CMU working with George Kantor on Simultaneous Localization and Mapping (SLAM) in agricultural environments. Prior to that, I worked with Simon Lucey and Laszlo Jeni on monocular 3D reconstruction. In my undergrad, I majored in Electrical Engineering at the University of Maryland - College Park.

Email  /  CV  /  Google Scholar  /  LinkedIn  /  Github

[Jul '24]  

[Apr '24]  

[Jan '24]  

[Jun '23]  

[Jun '23]  

[Jan '23]  

[Jan '23]  

[Jan '23]  

We extend the Gaussian splatting algorithms for two commonly used sonars (Imaging sonars and echosounders) and propose fusion algorithms that simultaneously utilize RGB camera data and sonar data.

We present a technique to fuse acoustic and optical measurements for 3D reconstruction. Our framework can reconstruct accurate high-resolution 3D surfaces from measurements captured over heavily-restricted baselines.

We propose a gradient-based method to estimate well-conditioned covariance matrices for estimation. We formulate the the learning procedure as a constrained bilevel optimization problem over factor graphs.

We propose a gradient-based method for learning observation models for robot state estimation with incremental non-differentiable optimizers in the loop. Our algorithm converges much quicker to model estimates that lead to solutions of higher quality compared to existing methods

We present a technique for dense 3D reconstruction of objects using an imaging sonar. We represent the geometry as a neural implicit function.

This paper presents a novel application of conditional Generative Adversarial Networks (cGANs) to train a model to produce noise-free sonar images.

We go beyond traditional parallelism of the A* algorithm and introduce RA* (Runahead A*) which performs speculative parallelism of A*.

We present an Augmented Lagrangian-based incremental constrained optimizer that views matrix operations as message passing over the Bayes tree.

RACOD is an algorithm/hardware co-design for mobile robot path planning. It consists of two main components: CODAcc, a hardware accelerator for collision detection; and RASExp, an algorithm extension for runahead path exploration.

We develop a computer vision method to size and track the growth rates of apple fruitlets. Fruitlets are sized using a combination of deep learning algorithms and are associated across days using a novel Attentional Graph Neural Network approach.

In this paper, we use semantics and environmental priors to construct accurate 3D maps of agricultural environments.

Source code