Vision-based autonomy for GPS-denied environments.
CorvidX builds the software that lets a drone map an unknown space and navigate it in real time — from nothing but a single camera and an IMU.
Commodity sensors, built to scale from simulation to the field. Dual-use, built in Canada.
A real replay: our drone exploring a photorealistic simulated environment it has never seen before. With no prior map, it avoids obstacles, identifies new frontiers to explore, and incrementally builds a SLAM-style 3D reconstruction of the space — using only a simulated camera and IMU. The full perception, mapping, planning, and control pipeline runs end-to-end in simulation today, with hardware bring-up on Jetson-class compute as the next step.
Canada's drone ecosystem is largely airframe-first: strong aircraft, with the autonomy that flies them treated as an afterthought or sourced elsewhere. But the hard part of operating in the real world — without GPS, without a datalink, without a map handed to you in advance — lives in software, not airframe. That intelligence layer is increasingly being carved out as a distinct capability rather than a feature of the aircraft.
CorvidX builds that layer.
Our stack runs entirely from a single monocular camera and an IMU — no LiDAR, no GPS, no prior map. Visual SLAM builds a 3D map of the environment as the drone moves through it; a planner reasons over that live map to avoid obstacles and seek out unexplored space; and a learning-based controller turns those decisions into motor commands directly.
Because it depends only on commodity sensors, the same autonomy that runs in simulation scales cleanly to low-cost field hardware. It is designed to run onboard, in real time, on Jetson-class compute.
Janahan Ramanan
FounderJanahan is a Senior ML Engineer and Technical Lead at Meta Reality Labs (Spatial AI), where he leads technical work on visual place-recognition, 3D mapping, and on-device computer vision for Ray-Ban Meta smart glasses. Before Meta he was a Technical Lead at Borealis AI, RBC's research institute. He is a published researcher (Time2Vec; OMEN, ICML 2021) and a patent holder, with an M.A.Sc. and B.Eng. in Engineering Science from the University of Toronto.
Sal Arif
Founding Product LeadSal is an AI product leader with 12+ years across capital markets and machine learning. He has held Associate Partner and Senior Principal Product Manager roles at QuantumBlack, AI by McKinsey, and was Group Product Manager at Borealis AI (Canada's largest ML lab), leading a team of product managers on frontier problems in quantitative finance. He has shipped AI products in financial firms that scaled to millions of transactions per second and traded over $20B in notional, and served as Director of Generative AI and AI & Analytics Product Management at RBC Capital Markets. An angel investor and venture partner, Sal holds an MMAI from Queen's University and an MSc in Financial Mathematics, with earlier roles as an investment banker and quant.