cd ~/

Year One at Magic Leap: A Retrospective

Reflecting on my first full year building mixed reality systems - what worked, what didn't, and what I've learned.

Evyatar Bluzer
2 min read

A year ago I joined Magic Leap knowing almost nothing about mixed reality. Today I lead the perception architecture for a device that's actually taking shape. Time for honest reflection.

What Went Well

Technical Foundation

We have working systems:

  • Visual-inertial odometry running at 60Hz with under 1mm tracking error
  • ToF depth sensor integrated and calibrated
  • Eye tracking prototype showing promising accuracy
  • Synthetic data pipeline generating training data

The architecture decisions from early 2017 are proving sound.

Team Growth

Started with 3 engineers, now have 15 across perception disciplines:

  • Strong technical hires who are already contributing
  • Culture of demo-driven development
  • Cross-functional collaboration improving

Synthetic Data Bet

The synthetic data team is already accelerating our ML development. Training eye tracking models on synthetic data is working better than I expected.

What Didn't Go Well

Schedule Optimism

We underestimated integration complexity. Individual components working doesn't mean the system works. Integration bugs consumed Q3-Q4.

Lesson: add 50% to integration estimates, minimum.

Hardware-Software Handoff

Communication gaps between hardware (optics, sensors) and software (algorithms) teams caused rework. Specs that seemed clear weren't.

Lesson: written specs aren't enough. Need working prototypes across the boundary.

Power Budget Overrun

We're 30% over power budget. Every subsystem optimized locally, but the sum exceeds the system constraint. Now doing painful trade-offs.

Lesson: track power continuously at system level, not just per-component.

Biggest Learnings

Perception is a System Problem

You can't optimize depth sensing without considering SLAM. You can't optimize SLAM without considering display latency. Everything connects.

Data is Strategy

The team that controls data generation controls algorithm development velocity. Synthetic data is strategic infrastructure.

Embedded is Different

Algorithms that work on desktop often can't be translated to embedded. Design for target platform from day one.

Calibration is Half the Work

Brilliant algorithms don't matter if calibration is wrong. Invest in calibration infrastructure early.

Looking Ahead to 2018

Priorities:

  1. Ship developer kit: Real devices in real developers' hands
  2. Power optimization: Get back within budget
  3. Synthetic data scaling: 10x current generation capacity
  4. Hand tracking: Add hands as input modality

It's going to be intense. But we're building something real. That's energizing.

Thanks to everyone on the team who made this year possible. Onward.

Comments