research

Papers.

Mostly one obsession: getting robots to work with less — less labelling, fewer sensors than you'd think you need, and honest numbers about what actually helps.

CloudDiffusion

IEEE SMC 2026 · published · arXiv:2606.16048

A LiDAR sweep sees maybe a twentieth of the scene. CloudDiffusion completes the rest — walls, ground, the parked car — in one denoising step, about 209 ms per frame where iterative diffusion methods take seconds.

drag to rotate

FIG. 1 — sparse sweep in, dense scene out, one step. 23–98× lower latency than iterative diffusion baselines on the same GPU.

Diffusion models usually buy quality with dozens of denoising iterations. The surprise in this work: the data mattered more than the model. Rebuilding the ground truth with anchor-based ICP cut the error about 16× with no model change, a compact multi-token VAE stayed stable at scene scale where earlier autoencoders collapsed, and a single x₀ step directly in coordinate space was enough to complete the scene. Co-first author with Chidera Agbasiere; code and weights are public.

HapticVLA

IROS 2026 · published

Tactile sensors are great in the lab and annoying on real robots. HapticVLA uses touch only during training — then distills it into a student that handles an egg with no tactile sensor on board, predicting the touch signal from what it sees.

switch policies — same egg, same gripper, no sensor either way

FIG. 2 — the plain policy squeezes blind past the crack line. the distilled student predicts the contact force (dashed) and settles in the safe band — sensor-free.

Two ingredients: a teacher trained with a safety-aware tactile reward that penalizes over-force and slip, and a distillation step that teaches a standard vision-language-action model to predict the tactile signal it no longer receives. Evaluated on real fragile-object tasks — eggs included. In review; numbers and write-up when the process allows.

HapticWAM

ICRA 2027 · submitted · arXiv:2609.23888

You know the glass is about to slip a beat before it does. HapticWAM is a world-action model that gives robots that beat: a teacher that feels the fingertip pads imagines the contact, and a distilled student runs on the real robot with no tactile sensors at all, 77% mean success on three contact-rich tasks, ahead of pi0.5 and Diffusion Policy.

FIG. 3 — hollow marks: what the model says is coming. filled marks: what the world then does. the gap is lead time — and high uncertainty visibly slows the motion.

The bet: a robot that anticipates contact doesn't need to sense it everywhere — rich touch can live in training and become prediction at deployment, with the model's own uncertainty acting as the speed governor. Early days; details when there's something worth defending.