Researcher

Erik Wijmans

Erik Wijmans is an AI researcher whose work spans embodied navigation, reinforcement learning, and large language model training. He earned bachelor's degrees in engineering physics and computer engineering in 2017 and completed a computer science PhD at Georgia Tech in 2022, when he shared the AAAI/ACM SIGAI Doctoral Dissertation Award. He coauthored Habitat and Habitat 2.0 for embodied agents, alongside DD-PPO and VER for scalable navigation and mobile-manipulation learning. He shared the 2023 ICLR Outstanding Paper Award for a study of maps learned by blind navigation agents, and his 2025 Cut Cross-Entropy paper reduced memory use in large-vocabulary language-model training. He works in AI research at Thinking Machines, where he scales Tinker and improves fine-tuning models.

Timeline

  1. 2017

    Two bachelor's degrees

    In 2017, he earns a BS in engineering physics at Juniata College, graduating magna cum laude, and a BS in computer engineering at Washington University in St. Louis, graduating summa cum laude.

  2. 2019

    Habitat for embodied AI

    In 2019, he coauthors Habitat, a platform for research in embodied AI that trains virtual agents in photorealistic 3D simulations. Habitat-Sim provides the simulator, and Habitat-API supports task definition, training, and benchmarking.

  3. 2020

    Scaling PointGoal navigation

    In 2020, he coauthors DD-PPO, a distributed reinforcement-learning method for virtual robots navigating Habitat-Sim. It scales training to 2.5 billion experience steps and reaches near-perfect PointGoal navigation in unseen environments.

  4. 2021

    Habitat 2.0 for home tasks

    In 2021, he coauthors Habitat 2.0, a physics-enabled simulator for virtual robots in interactive 3D environments, and the Home Assistant Benchmark for mobile-manipulation tasks such as tidying rooms, stocking groceries, and setting a table.

  5. 2022

    A PhD and dissertation award

    He completes a computer-science PhD at Georgia Tech in 2022 under Irfan Essa and Dhruv Batra and shares the AAAI/ACM SIGAI Doctoral Dissertation Award for his thesis. His 2022 paper VER scales on-policy reinforcement learning across many GPUs to train embodied-navigation and mobile-manipulation policies.

  6. 2023

    An ICLR Outstanding Paper Award

    In 2023, he shares the ICLR Outstanding Paper Award for Emergence of Maps in the Memories of Blind Navigation Agents. The study examines representations learned by artificial blind navigating agents.

  7. 2025

    Cut Cross-Entropy for LLMs

    In 2025, he coauthors Cut Your Losses in Large-Vocabulary Language Models, which introduces Cut Cross-Entropy to compute cross-entropy loss without materializing logits for all tokens into global memory.

  8. As of October 2026

    AI research at Thinking Machines

    In 2026, he joins Thinking Machines from Apple, where he works with Vladlen Koltun. At Thinking Machines, he works on scaling Tinker, the lab's fine-tuning service, and making fine-tuning models such as Inkling easier and more effective.

Career

  1. Since 2026Thinking Machines Lab

Education

  1. 2022Georgia Institute of TechnologyPhD in computer science
  2. 2017Juniata CollegeBS in engineering physics
  3. 2017Washington University in St. LouisBS in computer engineering

Sources and credits

Photo: We have no published photo of Erik Wijmans yet. Send us a free-licensed photo.

  1. Erik Wijmans wijmans.xyz. Checked 21 October 2026.
  2. Habitat: A Platform for Embodied AI Research openaccess.thecvf.com. Checked 21 October 2026.
  3. DD-PPO: Learning Near-Perfect PointGoal Navigators from 2.5 Billion Frames arxiv.org. Checked 21 October 2026.
  4. Habitat 2.0: Training Home Assistants to Rearrange their Habitat proceedings.neurips.cc. Checked 21 October 2026.
  5. AAAI/ACM SIGAI Doctoral Dissertation Award aaai.org. Checked 21 October 2026.
  6. VER: Scaling On-Policy RL Leads to the Emergence of Navigation in Embodied Rearrangement arxiv.org. Checked 21 October 2026.
  7. Announcing the ICLR 2023 Outstanding Paper Award Recipients blog.iclr.cc. Checked 21 October 2026.
  8. Cut Your Losses in Large-Vocabulary Language Models proceedings.iclr.cc. Checked 21 October 2026.
  9. Meta’s loss is Thinking Machines’ gain techcrunch.com. Checked 21 October 2026.