We study how autonomous systems perceive, localize, and plan in dense, heterogeneous traffic — environments where motorcycles, paratransit vehicles, and unstructured road behavior define the operating reality.
MixedDrive Lab is an independent research initiative investigating perception, localization, and decision-making for autonomous systems operating in mixed-traffic environments — characterized by dense two-wheeler populations, weak lane discipline, and irregular vehicle geometries.
The dominant body of autonomous-driving research has been shaped by highway-centric and structured-urban assumptions from North American, European, and Chinese contexts. Southeast Asian traffic violates most of those assumptions. Our work characterizes where existing methods fail, and develops approaches that remain robust under these conditions.
The lab operates in a deliberate, publication-driven cadence: empirical work first, claims second.
Empirical characterization of failure modes in state-of-the-art visual SLAM systems (ORB-SLAM3, VINS-Fusion) when operated in motorcycle-dominated, dynamic-object-rich scenes. Toward feature-selection strategies that remain stable under aggressive scene change.
Visual-inertial odometry tailored to motorcycle-class platforms, where roll dynamics, vibration profiles, and mounting geometries differ substantially from automotive assumptions baked into existing VI-SLAM stacks.
Game-theoretic motion planning for vehicles navigating around paratransit and informal road users — agents whose intent is poorly captured by classical predict-then-plan pipelines. Future work, contingent on perception-stack maturity.
The lab is open to correspondence with academic researchers, graduate students, and engineers working on related problems in perception, localization, or motion planning under unstructured-traffic conditions.
[email protected]