Ryan Rahman

Robot learning researcher working on long-horizon manipulation, post-training, and force-aware manipulation.

Founding Research Engineer at Preload · Accelerated MASc researcher at Waterloo.

Research Email

Ryan Rahman smiling while holding his black cat

Founding Research Engineer

Preload · San Francisco

University of Waterloo

Accelerated MASc · Mechatronics & robotics

Selected work

All research

Mean LIBERO success: Dream-7B 82.2%, matched Qwen2.5-7B 76.3%. Single-seed simulation results.

Diffusion-trained backbones for flow-matching VLA policies

A controlled backbone swap improves mean LIBERO success by 5.9 points, with the largest gains on long-horizon tasks. Single-seed, simulation-only study.

Read the preprint

The bimanual hardware for tactile-aware folding.

Origami-MoH

Reactive, multi-horizon flow policies and progressive tactile fusion for long-horizon folding on a 65-DoF bimanual platform.

Explore the proposal

Rocky

A five-legged robot designed end to end, from a Fusion 360 model to an analytic wave gait and a learned PPO walking policy.

Explore the build Code