Opens in a new tab
BLOG

Thinking out loud about engineering and research.

Essays and tutorials on machine learning, brain-computer interfaces, robotics, software, and the messy gap between academic research and production systems. New posts when I have something to say.

Featured · ML Research

The Reproducibility Problem in Deep Reinforcement Learning: What We Can Actually Do About It

After three years of trying to reproduce results from top-tier reinforcement learning papers, and failing more often than I'd like to admit, I've come to believe the field's reproducibility crisis goes deeper than most people acknowledge. Here's what I think is going wrong, and what a more honest culture of experimentation could look like.

E
Emre ARI
March 18, 202614 min read
All Writing

Recent articles

AllML ResearchEngineeringRoboticsOpinionTutorials
Robotics

Why Sim-to-Real Transfer Still Fails in the Interesting Cases

The gap between simulation and real-world performance isn't only about physics accuracy. Here's what I learned from two years of trying to close it on a collaborative arm.

Feb 4, 202611 minRead
Engineering

Writing C++ for Robotics: Lessons from Building Kepler

Building a motion planning library that researchers actually want to use is harder than getting the algorithms right. Here are the software engineering decisions that helped three research groups adopt Kepler.

Jan 20, 20269 minRead
ML Research

The Case for Smaller, Better-Evaluated Models

The race toward scale has produced impressive benchmarks and underwhelming deployment stories. A contrarian argument for why the next decade of ML should go deep rather than large.

Jan 6, 202613 minRead
ML Research

What I Learned from Six Months of Benchmarking Energy Forecasting Models

A deep dive into temporal distribution shift, evaluation methodology, and why the model that tops the leaderboard almost never wins in production. Lessons from the RTX project.

Oct 10, 202510 minRead
Secret Link