Ian Abraham

Research profile

Research summary

Ian Abraham is a robotics researcher at The University of Sydney whose work sits at the intersection of robot learning and control theory; OpenAlex field data splits his profile evenly (0.5/0.5) between robotics (robot learning) and robot learning (robotics and control theory). His indexed record shows an h-index of 15 and an i10-index of 20, with output concentrated in recent years: 33 works in the last three years, peaking at 13 indexed works in both 2025 and 2026, which drives a maximum activity score. Topic modelling ranks reinforcement learning in robotics first (42 works), followed by robotic path planning algorithms (22), robot manipulation and learning (15), optimization and search problems (12), and adaptive dynamic programming control (10) — a spread that yields a low focus score. Evolve has reviewed two of his 2026 arXiv papers. 'Accelerating Visual Policy Learning with Sampling-Based Model Predictive Control' (last author) pairs sampling-based MPC with first-order policy gradients for single-GPU visual policy learning and zero-shot Go2 transfer; Evolve calls it promising but statistically thin. 'Ergodic Control and Controlled Diffusion for Robot Learning: Review and Tutorial' (middle author) unifies ergodic control and controlled diffusion for robot data collection with rigorous mathematics and practical recipes, though empirical benchmarks remain limited. He works within a compact network of 12 distinct recent co-authors.

Research keywords

Highlights

• Single-GPU visual policy learning with sampling-based MPC

  In a 2026 arXiv paper (last author), SGPS combines sampling-based model predictive control (MPPI) with first-order policy gradients and differentiable simulation to train visual policies on a single GPU. Evolve's verdict: promising approach, though supporting statistics are thin; reproducibility readiness rated medium.

• Zero-shot sim-to-real transfer to the Go2 quadruped

  The same paper reports zero-shot transfer of learned visual policies to the Go2 legged robot, positioning sampling-based MPC as an accelerator for sim-to-real locomotion learning; Evolve notes the empirical evidence base remains thin, with keywords spanning legged locomotion, differentiable simulation, and MPPI.

• Tutorial unifying ergodic control and controlled diffusion

  A 2026 arXiv review/tutorial (middle author) presents ergodic control and controlled diffusion as a unified framework for robot learning and data collection, covering coverage control and stochastic optimal control with rigorous mathematics and practical recipes; Evolve rates its reproducibility readiness high while noting limited empirical benchmarks.

Timeline

1. 2013First indexed works appear in OpenAlex: 1 publication in 2013, with 1 each in 2014 and 2015.
2. 2017Early growth phase: 3 indexed works in 2017, 2 in 2018, and 3 in 2019.
3. 2020Output accelerates: 6 indexed works in 2020, then 5 in each of 2021 and 2022.
4. 2023Sustained productivity: 9 works in 2023 and 7 in 2024.
5. 2025Peak activity: 13 indexed works in 2025, matched again in 2026, with 33 works over the last three years.
6. 2026Two 2026 arXiv papers enter Evolve's catalog: the SGPS visual-policy-learning paper (last author) and the ergodic control and controlled diffusion tutorial (middle author).

Papers in the Evolve catalog

• 2026-09-17

  Accelerating Visual Policy Learning with Sampling-Based Model Predictive Control

• 2026-09-09

  Ergodic Control and Controlled Diffusion for Robot Learning: Review and Tutorial

Frequent co-authors