MixedDrive Lab
Autonomous systems for emerging contexts.
An independent research initiative based in Bandung, Indonesia, working on perception, localization, and behavior modeling for autonomous systems operating in environments that don't match the assumptions baked into mainstream AV stacks — Southeast Asian mixed traffic, paratransit, and adjacent autonomy domains.
What we work on
Visual SLAM in motorcycle-dominated mixed traffic. Most production VSLAM systems are evaluated on KITTI-style datasets — disciplined lane-following, low pedestrian density, predictable vehicle scale. Bandung traffic isn't that. We characterize where ORB-SLAM3, VINS-Fusion, and similar pipelines fail in this setting and develop feature-selection and behavior-aware methods that hold up.
Mixed-traffic-aware feature selection. Building on the failure-mode analysis: scoring functions that combine gradient strength, temporal stability, and traffic-aware penalties, validated against ORB-SLAM3 baselines on Bandung clips and CARLA scenarios.
Underwater autonomy. Collaborative work with ITB on tethered micro-ROV relative localization — observability analysis (PCRLB), learned inertial odometry, and low-cost PVDF hydrophone arrays.
Multi-agent / paratransit behavior. Game-theoretic motion planning for interaction with informal transit (angkot, ojek) — the agents that don't fit into standard intelligent-vehicle taxonomies. (Phase 2, post-2028.)
Current focus

• Empirical characterization of VSLAM failure modes in motorcycle-dominated mixed traffic. Paper in preparation, ICSET 2026.
• Mixed-Traffic-Aware Feature Selection for Robust Visual SLAM. Paper in development, IEEE IV 2028 target.
• zenbook-s16-ubuntu-setup — open-source: battle-tested Ubuntu 24.04 setup for AMD Ryzen AI laptops with amdxdna NPU compatibility fixes.
  Selected publications
  *Full list: * Google Scholar * · IEEE Author ID: * 38469211300
  Note: this section will be updated post-Paper #1 submission. Pre-2026 publications listed below are from prior work and are kept for context.
• (placeholder — pick 3–5 strongest from the 2013–2022 IEEE/Medwell record and replace this list before publishing)
• M. V. G. Aziz, "[Selected pre-Lab publication 1 — title]", [Venue], [Year].
• M. V. G. Aziz, "[Selected pre-Lab publication 2 — title]", [Venue], [Year].
• M. V. G. Aziz, "[Selected pre-Lab publication 3 — title]", [Venue], [Year].
  Research approach
  We treat reproducibility as engineering, not a virtue signal. Every paper from MixedDrive Lab ships with the code, the dataset (or a public subset), and the setup scripts needed to re-run the experiments. Solo research means decisions move fast — and they're documented, so future collaborators (or future you) can audit the path.
  Lab status
  MixedDrive Lab currently operates as a solo research program, founded in 2026. The "Lab" designation reflects the program structure and intent rather than its current institutional infrastructure. Long-term plan: grow into a multi-researcher team and partnerships with regional academic institutions.
  We're not currently looking for collaborators (the bandwidth isn't there yet), but the work is open and the code is liberally licensed. If something here intersects with what you're working on, drop a line.
  Contact
• Web: mixeddrivelab.org
• Email: vicky@mixeddrivelab.org
• Founder: Dr. Vicky Ghani (Institut Teknologi Bandung, 2020)
  
  Bandung, Indonesia · Independent research · MIT-licensed code unless otherwise stated