AI Wiki

Explore AI Wiki

Citation and evidence

Jeff Hawke

14 min full readUpdated 31 references

This article's verification

Report a problem with this article

More

Use this article

Raw MarkdownExplore connections

Improve this page

Suggest editRevision historyDiscussion

Browse categories

Autonomous VehiclesPeopleRoboticsWorld Models

Cite this article

Jeff Hawke (published as Jeffrey Hawke) is a New Zealand robotics and machine learning engineer who is the co-founder and chief technology officer of Odyssey, the world model lab he started with Oliver Cameron in 2023 [1][9][10]. Before Odyssey he worked at the British autonomous driving company Wayve from its early days until 2023, described by an early investor as its founding engineer and, by 2022, as its vice president of technology; he is a co-author of Wayve's 2018 paper Learning to Drive in a Day, which its authors presented as the first application of deep reinforcement learning to autonomous driving, and the first-listed author of its 2019 paper on end-to-end urban driving with conditional imitation learning [5][6][8][21][23]. He completed a DPhil at the University of Oxford's mobile robotics group, where his research concerned perception systems for self-driving cars [1][3][4].

At Odyssey he is credited under "Leadership" on Starchild-1 and Agora-1, is an author of the PROWL and CaliBench papers, wrote the company's announcement of PROWL-1, co-wrote the announcement of Odyssey-3 with Cameron, and presented Odyssey's research at the RAAIS 2026 summit in London [12][14][15][16][17][19][27].

Early life and education

Hawke was raised in Mt Albert, a suburb of Auckland, and graduated from the University of Auckland in 2009 in mechatronics engineering and computer science; the Oxford Robotics Institute records the degrees as a Bachelor of Engineering (Hons) and a Bachelor of Science [3][5]. His first graduate job, in his own account to the New Zealand Herald, was developing algorithms for an autonomous forklift at a start-up, Inro Technologies, which the Herald said entered liquidation a couple of years later [5]. The Oxford Robotics Institute's biography likewise describes him working as a robotics engineer at a New Zealand start-up developing autonomous forklifts, and completing internships at Rethink Robotics and Willow Garage [3].

He then completed a Master of Science in Mechanical Engineering at the Georgia Institute of Technology in 2012 on a Fulbright Scholarship, working in the Healthcare Robotics Lab on control systems and motion planning for robotic manipulation with whole-arm tactile sensing [3]. He is a co-author on two Georgia Tech papers from that period, on whole-arm tactile sensing during robotic assistance (ICORR 2013) and on interleaving planning and control for haptically guided reaching (Humanoids 2014), and on a 2014 International Journal of Robotics Research article describing the Velo gripper, a single-actuator design for enveloping, parallel and fingertip grasps [3].

In 2013 Hawke moved to the University of Oxford as a DPhil student and member of University College, working in the Mobile Robotics Group, whose alumni pages are now kept by the Oxford Robotics Institute, on perception for the group's autonomous car project, in particular on how scene information can improve object detection and support lifelong learning in perception systems [3][4]. His Oxford papers with Ingmar Posner and colleagues include "Learning on the Job: Improving Robot Perception Through Experience" (NIPS 2014 workshop on autonomously learning robots), "Wrong Today, Right Tomorrow: Experience-Based Classification for Robot Perception" (Field and Service Robotics, 2015) and, as first author, "What Makes a Place? Building Bespoke Place Dependent Object Detectors for Robotics" (IROS 2017), which fits lightweight pedestrian detectors to the particular places a robot repeatedly drives through and reports a sizeable gain over a state-of-the-art general detector [3][26]. Air Street Capital's 2026 speaker profile states that he completed his doctorate at Oxford, and Odyssey's leadership page lists him as "Dr. Jeff Hawke" [1][11].

A 2020 conference speaker profile adds that before Wayve he worked as a strategy consultant at the Boston Consulting Group [7].

Wayve

Hawke is listed as the second author, after Kendall, of the July 2018 paper Learning to Drive in a Day, which reported what its authors called "the first application of deep reinforcement learning to autonomous driving": a policy for lane following learned from randomly initialised parameters in a handful of on-vehicle episodes, using a single monocular camera and a reward equal to the distance travelled before the safety driver intervened [21]. The paper was later presented at ICRA 2019 [21]. Nathan Benaich of Air Street Capital, who disclosed being an angel investor in Wayve, described Hawke in 2024 as "the founding engineer at Wayve", who "pioneered the use of end-to-end models to drive cars on complex roads" and "played a leading role in building the company" [8].

His title at Wayve appears differently in sources from different years. A speaker profile for an Informa transport conference in 2020 lists him as Chief Strategy Officer and says he worked "across technology and strategy" [7]; an Oxford seminar listing from February 2020 says he worked "across tech and business strategy" [4]; MIT Technology Review quoted him in September 2022 as "Wayve's vice president of technology" [6]; and the Air Street and RAAIS profiles from 2026 describe him as having been VP Technology, credited with helping "pioneer visual policy learning" and end-to-end learning approaches for autonomy [1][2].

Hawke's Wayve-era research followed the company's move from lane following to urban driving. He was the first-listed author of "Urban Driving with Conditional Imitation Learning" (arXiv November 2019, ICRA 2020), which presented an end-to-end conditional imitation learning system combining lateral and longitudinal control on a real vehicle following user-defined urban routes with simple traffic, learned from human driving demonstrations and built on learned representations of semantics, geometry and motion [23]. He was a co-author of "Learning to Drive from Simulation without Real World Labels" (ICRA 2019), which transferred a vision-based lane-following policy from simulation to a rural road using image-to-image translation, a sim-to-real transfer result evaluated in closed loop on a real car [22], and of FIERY (ICCV 2021), a probabilistic bird's-eye-view future instance prediction model trained end to end from surround monocular cameras without HD maps [24]. In August 2021 he was lead author of the position paper "Reimagining an autonomous vehicle", written with Haibo E, Vijay Badrinarayanan and Alex Kendall, which argued that the industry's designs were "rooted in decisions made a decade ago" and set out what the authors called AV2.0, a recipe for driving with machine learning together with grand challenges for research [25].

He also acted as a public voice for the company. In February 2020 he gave a seminar titled "Learned Urban Driving" in the seminar series of Oxford's AIMS doctoral training centre, describing how Wayve learned driving policies "without relying on high definition maps or hand-coded, rule-based planners" [4]. When Wayve showed in 2022 that one model could drive both a passenger car and a delivery van, he told MIT Technology Review, "It's like when you go somewhere new and get a rental car, you can still drive." The same piece closed with his caution: "I don't want to diminish the scale of the challenge ahead of us. The AV industry teaches you humility" [6].

According to the New Zealand Herald, he left Wayve in 2023 for what was then a stealth start-up [5].

Odyssey

Hawke co-founded Odyssey with Oliver Cameron, the former Voyage chief executive and Cruise product executive, in 2023; Cameron wrote in June 2026 that "when @jeffrey_hawke and I started @odysseyml in 2023, we believed general world models would become a new class of foundation model" [9][10]. Hawke is the company's chief technology officer [1][9][11]. The Herald reported that the two founders arrived at the same idea from self-driving: that the world simulation technology built to help cars drive could be decoupled from that domain and used to learn the physics of the whole world for robotics, games and interactive video [5]. Air Street Capital, which announced in July 2024 that it had invested in Odyssey's $8 million first round, wrote that both co-founders "come from the self-driving world" [8].

Odyssey's staff credits track his involvement across the model line. He appears among the technical staff on the May 2025 research preview of Odyssey-1 and in the team lists for Odyssey-2 and Odyssey-2 Max, and is credited under "Leadership", with Cameron, on both Starchild-1 (May 2026), a real-time audio-video world model, and Agora-1 (May 2026), a multi-agent world model [17][27][28][30][31]. Starchild-1's technical report lists him as "Jeffrey Hawke" under leadership, and Odyssey published a research panel, "The making of Starchild-1", in which Hawke, Jenny Seidenschwarz and Vighnesh Birodkar discuss the model's core ideas and technical challenges [18][29].

His two bylined pieces on the Odyssey blog set out the lab's research position. In "On the Origin of Species (of World Models)" (27 February 2026) he argued that the term world model was "being overloaded" and proposed a four-way taxonomy: world models proper, which "learn how the world evolves" (he named Odyssey and DeepMind's Genie as examples and JEPA as a specific interpretation); spatial intelligence, which learns how the world appears (World Labs); behaviour models, which learn how to act within a world (he placed Wayve's driver and Physical Intelligence here); and "proxy world models", the internal representations of language models [13]. In "Introducing PROWL-1: Learning Through Discovery" (12 May 2026) he described a reinforcement learning agent trained to explore a Minecraft environment and find failures in a world model initialised with Diffusion Forcing, with a prioritized adversarial trajectory buffer turning those failures into a curriculum; he framed the goal as giving world models the kind of "scalable feedback loops" that language models already have [14]. The corresponding paper, "PROWL: Prioritized Regret-Driven Optimization for World Model Learning" (arXiv 2605.18803), lists him as an author with Odyssey affiliation alongside Ahmet H. Güzel, Jenny Seidenschwarz, Benjamin Graham, Jonathan Sadeghi and Ilija Bogunovic [15].

He is also the last-listed author of CaliBench (arXiv 2608.16829), a benchmark testing whether video world models reproduce the correct distribution of physical outcomes in settings with a known reference distribution, such as Galton boards and dice; the paper was accepted at Transactions on Machine Learning Research in 2026 [16]. On 15 September 2026 he and Cameron jointly bylined "Introducing Odyssey-3: A General-Purpose Physical Intelligence", which opens by recalling that the two "began working on autonomous vehicles and robotics in the 2010s" and founded Odyssey "around the belief that world models could provide a legitimate technical path" to general-purpose physical intelligence [12].

In public, Hawke has been the company's main technical explainer. He was interviewed on The Data Exchange podcast in March 2026, describing Odyssey's models as producing "a continuous stream of intelligent pixels that you can interact with", saying the majority of training data is "large-scale public video", and characterising the period as "the GPT-2 era of world models" [20]. On 12 June 2026 he spoke at the tenth Research and Applied AI Summit (RAAIS) in London, in a talk Odyssey titled "A missing form of intelligence" [1][19]. He told the Herald in July 2026, after Odyssey's $310 million Series B, that about half of the company's roughly 35 staff were in King's Cross, London, and the other half in Palo Alto [5].

Selected publications

Author lists follow the arXiv listing, the Oxford Robotics Institute publication page or the DOI record; the name form on all of them is "Jeffrey Hawke" or "J. Hawke".

YearTitleVenuePosition and notable co-authors
2013Whole-arm tactile sensing for beneficial and acceptable contact during robotic assistanceICORR 2013Co-author, with Charles C. Kemp [3]
2014The Velo gripper: A versatile single-actuator design for enveloping, parallel and fingertip graspsInternational Journal of Robotics Research 33(5)Co-author, with Matei Ciocarlie [3]
2014Learning on the Job: Improving Robot Perception Through ExperienceNIPS 2014 workshopSecond author, with Corina Gurau and Ingmar Posner [3]
2015Wrong Today, Right Tomorrow: Experience-Based Classification for Robot PerceptionField and Service Robotics 2015First author, with Ingmar Posner [3]
2017What Makes a Place? Building Bespoke Place Dependent Object Detectors for RoboticsIROS 2017First author, with Alex Bewley and Ingmar Posner [26]
2018Learning to Drive in a DayarXiv; ICRA 2019Second author, with Alex Kendall and Amar Shah [21]
2018Learning to Drive from Simulation without Real World LabelsarXiv; ICRA 2019Co-author, with Alex Bewley and Alex Kendall [22]
2019Urban Driving with Conditional Imitation LearningarXiv; ICRA 2020First-listed author, with Richard Shen, Corina Gurau, Amar Shah and Alex Kendall [23]
2021FIERY: Future Instance Prediction in Bird's-Eye View from Surround Monocular CamerasICCV 2021Co-author, with Anthony Hu, Roberto Cipolla and Alex Kendall [24]
2021Reimagining an autonomous vehiclearXivFirst author, with Vijay Badrinarayanan and Alex Kendall [25]
2026PROWL: Prioritized Regret-Driven Optimization for World Model LearningarXivCo-author, with Ahmet H. Güzel and Ilija Bogunovic [15]
2026CaliBench: Are the Stochastic Dynamics of Video World Models Physically Calibrated?TMLR 2026Last author, with Jonathan Sadeghi and Jenny Seidenschwarz [16]
2026Starchild-1: A real-time multimodal world modelOdyssey technical reportCredited under "Leadership" with Oliver Cameron [29]

Expanded article table

References

  1. ^1 ^2 ^3 ^4 ^5 ^6Air Street Press and Nathan Benaich, "Announcing Jeff Hawke (Odyssey) at RAAIS 2026", 10 March 2026. press.airstreet.com/...jeff-hawke-odyssey-raais-2026
  2. ^RAAIS, "Jeff Hawke, Odyssey", speaker profile, RAAIS 2026. raais.co/speakers-2026-jeff-hawke-odyssey
  3. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11Oxford Robotics Institute, "People: Jeff Hawke (alumni)", University of Oxford. ori.ox.ac.uk/...jeff-hawke
  4. ^1 ^2 ^3 ^4Oxford AIMS CDT, "Jeffrey Hawke Seminar: Learned Urban Driving", Department of Engineering Science, University of Oxford, 21 February 2020. aims.robots.ox.ac.uk/...jeffrey-hawke-seminar
  5. ^1 ^2 ^3 ^4 ^5 ^6Chris Keall, "Odyssey co-founder Jeff Hawke's journey from forklifts to a real-world AI start-up that's just raised $529m", The New Zealand Herald, 10 July 2026. nzherald.co.nz/...CUL34N2HCZEBPJCCFVLCNPDDIM
  6. ^1 ^2 ^3Will Douglas Heaven, "This startup's AI is smart enough to drive different types of vehicles", MIT Technology Review, 26 September 2022. technologyreview.com/...ehicles-cars-vans-scale-up
  7. ^1 ^2Informa Connect, "Jeff Hawke, Chief Strategy Officer at Wayve", Smart Transportation and Mobility speaker profile, 2020. tmt.knect365.com/...jeff-hawke
  8. ^1 ^2 ^3Air Street Press and Nathan Benaich, "Our investment in Odyssey", 8 July 2024. press.airstreet.com/...our-investment-in-odyssey
  9. ^1 ^2 ^3Julie Bort, "World model maker Odyssey nabs $1.45B valuation backed by Amazon and other big names", TechCrunch, 17 June 2026. techcrunch.com/...ed-by-amazon-and-other-big-names
  10. ^1 ^2Oliver Cameron (@olivercameron), post on X, 17 June 2026. x.com/...2067279298537336860
  11. ^1 ^2Odyssey, "Careers: Meet our leadership team". odyssey.systems/careers
  12. ^1 ^2Oliver Cameron and Jeff Hawke, "Introducing Odyssey-3: A General-Purpose Physical Intelligence", Odyssey, 15 September 2026. odyssey.systems/introducing-odyssey-3
  13. ^Jeff Hawke, "On the Origin of Species (of World Models)", Odyssey, 27 February 2026. odyssey.systems/...igin-of-species-of-world-models
  14. ^1 ^2Jeff Hawke, "Introducing PROWL-1: Learning Through Discovery", Odyssey, 12 May 2026. odyssey.systems/introducing-prowl-1
  15. ^1 ^2 ^3Ahmet H. Güzel, Jenny Seidenschwarz, Benjamin Graham, Jonathan Sadeghi, Jeffrey Hawke and Ilija Bogunovic, "PROWL: Prioritized Regret-Driven Optimization for World Model Learning", arXiv:2605.18803, May 2026. arxiv.org/...2605.18803
  16. ^1 ^2 ^3Jonathan Sadeghi, Jenny Seidenschwarz, Jesse Allardice, Sirish Srinivasan, Benjamin Graham and Jeffrey Hawke, "CaliBench: Are the Stochastic Dynamics of Video World Models Physically Calibrated?", arXiv:2608.16829, Transactions on Machine Learning Research, 2026. arxiv.org/...2608.16829
  17. ^1 ^2Oliver Cameron, "Introducing Starchild-1", Odyssey, 17 May 2026. odyssey.systems/introducing-starchild-1
  18. ^Odyssey, "The making of Starchild-1", research panel. odyssey.systems/the-making-of-starchild-1
  19. ^1 ^2Odyssey, "A missing form of intelligence: Odyssey's CTO Jeff Hawke presents on Odyssey's research at RAAIS 2026". odyssey.systems/raais-2026
  20. ^Ben Lorica, "World Models Are Here, But It's Still the GPT-2 Phase", The Data Exchange, 19 March 2026. thedataexchange.media/jeff-hawke-odyssey
  21. ^1 ^2 ^3 ^4Alex Kendall, Jeffrey Hawke, David Janz, Przemyslaw Mazur, Daniele Reda, John-Mark Allen, Vinh-Dieu Lam, Alex Bewley and Amar Shah, "Learning to Drive in a Day", arXiv:1807.00412, July 2018; ICRA 2019, doi:10.1109/ICRA.2019.8793742. arxiv.org/...1807.00412
  22. ^1 ^2Alex Bewley, Jessica Rigley, Yuxuan Liu, Jeffrey Hawke, Richard Shen, Vinh-Dieu Lam and Alex Kendall, "Learning to Drive from Simulation without Real World Labels", arXiv:1812.03823, December 2018; ICRA 2019, doi:10.1109/ICRA.2019.8793668. arxiv.org/...1812.03823
  23. ^1 ^2 ^3Jeffrey Hawke, Richard Shen, Corina Gurau, Siddharth Sharma, Daniele Reda, Nikolay Nikolov, Przemyslaw Mazur, Sean Micklethwaite, Nicolas Griffiths, Amar Shah and Alex Kendall, "Urban Driving with Conditional Imitation Learning", arXiv:1912.00177, November 2019; ICRA 2020, doi:10.1109/ICRA40945.2020.9197408. arxiv.org/...1912.00177
  24. ^1 ^2Anthony Hu, Zak Murez, Nikhil Mohan, Sofía Dudas, Jeffrey Hawke, Vijay Badrinarayanan, Roberto Cipolla and Alex Kendall, "FIERY: Future Instance Prediction in Bird's-Eye View from Surround Monocular Cameras", ICCV 2021, arXiv:2104.10490. arxiv.org/...2104.10490
  25. ^1 ^2Jeffrey Hawke, Haibo E, Vijay Badrinarayanan and Alex Kendall, "Reimagining an autonomous vehicle", arXiv:2108.05805, August 2021. arxiv.org/...2108.05805
  26. ^1 ^2Jeffrey Hawke, Alex Bewley and Ingmar Posner, "What Makes a Place? Building Bespoke Place Dependent Object Detectors for Robotics", IROS 2017, arXiv:1708.02330. arxiv.org/...1708.02330
  27. ^1 ^2Oliver Cameron, "Introducing Agora-1", Odyssey, 18 May 2026. odyssey.systems/introducing-agora-1
  28. ^Oliver Cameron, "Introducing Odyssey-1", Odyssey, 28 May 2025. odyssey.systems/introducing-odyssey-1
  29. ^1 ^2Team Odyssey, "Starchild-1: A real-time multimodal world model", technical report, 2026 (contributor list: "Leadership: Jeffrey Hawke, Oliver Cameron"). starchild.odyssey.ml/starchild-1.pdf
  30. ^Oliver Cameron, "Introducing Odyssey-2: A General-Purpose World Model", Odyssey. odyssey.systems/introducing-odyssey-2
  31. ^Oliver Cameron, "Introducing Odyssey-2 Max: Scaled World Simulation", Odyssey. odyssey.systems/introducing-odyssey-2-max

Improve this article

Add missing citations, update stale details, or suggest a clearer explanation. Every suggestion is reviewed for sourcing before it goes live.

1 revision · v2 · 2,781 words · full history

Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify

Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here

Reviewer note: Adversarial verification 2026-09-16 (cluster V7): every date, degree, title, paper and pronoun checked against its cited source; 8 arXiv author lists and 4 DOIs verified; 5 minor defects fixed

Cite this page: AI Wiki. "Jeff Hawke." aiwiki.ai, updated 16 Sept 2026, fact-checked 16 Sept 2026. CC BY 4.0. https://aiwiki.ai/wiki/jeff_hawke

Suggest edit