I'm focusing on starting 2027 by getting more actively connected to research in a variety of fields for the year.
This post is the offspring of Research Areas for 2025 which overlaps quite a bit, although I don't have a great record for how well I stuck to it. Coincidentally, there's a good amount of overlap.
Themes¶
Unified Models¶
Goal: Unified Models Across Control, Planning and State Estimation, derived from data
Ideally: start with a hypothesis for the model, identify the key parameters and missing features for a full complexity model. Then, identify simplification stages for partial complexity. These models can then be used to understand how the vehicle would react to control inputs (e.g. for a MPC step), how we expect the system to evolve (e.g. for a Kalman Filter) or how to convert plans from configuration space to metric space (e.g. for plan search).
Robust Architectures¶
How can one design robust architectures for localization, control or planning?
Computer Vision Patterns¶
Identify key architectures and patterns for computer vision tasks: instance segmentation, background segmentation, place recognition.
ML Best Practices¶
How can I bring those to my ML work as well as bring them to other areas of data-driven work?
This also extends to running conceptually larger models efficiently and with low latency in lower resource environments (e.g. neural network on CPU or small GPU).
Topics¶
Control¶
Re-cover a foundation of controls concepts (e.g. Steven Brunton's Control Bootcamp ). Related to the practices section from the 2025 post, my goal is to implement various pieces of the logic myself to confirm a better level of understanding for my self-study.
An extension area for me here is control barrier functions and trying to fit them into how they can positively interact with other parts of the system: for example, planning knowing for a controller with a known barrier function and integrating black-box or gray-box control algorithms with
Planning¶
Explore different types of planning algorithms: - Optimization - Search
State Estimation¶
I'd like to include learning about general architectures and backends while also exploring how different sensor modalities are integrated. - Camera - IMU - Encoders - Sonar - Lidar - WiFi / Beacons
Celestial Navigation¶
I'd like to loop back to the paper (paper, previous post). It sparked some new ideas for me and generally renewed my interest in learning celestial navigation as a skill
Drones¶
Application areas: - Visual Odometry - Visual Inertial Odometry - VI SLAM w/ Factor Graphs
Underwater Autonomy¶
An additional topic area that I see as logically adjacent to drones is underwater autonomous vehicles. Once I've taken some time to explore through the visual odometry for flying drones, I'd also like to spend some time exploring similar challenges for underwater vehicles.
Algorithm Design¶
Some key phrases to continue from: - anytime algorithms, resumable algorithms and checkpointable algorithms - cache-aware algorithms and cache-oblivious algorithms - structured concurrency
Algorithmic optimization for lower latency especially will continue to leak into my interests, especially if it can cross into the world where SIMD or understanding of modern CPU architectures (pipelines, cache) offers insight.