I'm thrilled to share that I've accepted a PhD position at Leiden University working on metamaterials for adaptive optics, starting later this year! ✨🔭 🇳🇱.

RT @dgdvision: We're featured in the Australian Financial Review today: Aussie experts can stop your smart-device cameras spying on you htt….

🐍Python update!. Our privacy reconstruction challenge now contains a python quick-start! All the instructions and contributions from others can be found at Project page:

Do you think you can crack this code?. We have made a discussion forum for our privacy reconstruction challenge! All the instructions and contributions from others can be found at Project page:

I am very happy that my work on privacy-preserving cameras has been seen internationally - das ist gut!. Project page:

Another video of what a privacy-preserving camera might "see"!. Project page: Media release:

Ever wonder what our privacy preserving camera actually sees? Here's some simulated "fingerprints" of scenes that encode location features without digitizing the full image. Suggestions for cool videos to try welcome!.Project page:

Want a robot that can't take photos of you? Our work on privacy preserving cameras was featured on @9NewsSyd yesterday It was great to work with @dgdvision @nikoSuenderhauf and Peter Corke, building off my thesis!.Project page:

RT @NC_Renic: Lecturers: “I’ve got some great comedic references in store for you this semester, I assume you’re all familiar with The Simp….

RT @ben_finegold: This @AustralianOpen semi has been seen before in a song. “I’m a joker…. I’m a sinner….” Etc.

Proof (by Thorlabs): g=10 QED

Me, after 5 mins of doing manual labour around the house: damn why isn't there a robot for this.

Trying my hand at this AI generation for a poster. perhaps the humans are robots? #AGI #stablediffusion .prompt: Robot vacuum cleaner taking a photo of a human in the bathroom. Picture from the robot perspective

7/ Ultimately, we hope this work will illicit wider discussions within the research community.

6/ Scrambling in the analogue domain means the sources of data is not known. For the result below, an attacker would have to recreate the image using only the values at the crosses in the bottom row, without knowing the locations.

5/ We also present a case study from a portion of my honours thesis: localisation. In simulation, we show that it is possible to obtain localisation accuracy compared to SIFT while collecting 10^-3 times less data.

4/ We propose inherently privacy-preserving vision: where the camera is specialised, processing is shifted to the optical/analogue domain, information is destroyed in processing and finally obfuscated. Ideally, this results in brute force attacks as the only remaining option.

3/ However, it is not an area of focus in robotics, with only 0.5% of papers mentioning it. Prior works generally are open to digital attack and/or do not necessarily prevent reconstruction. The one exception, Key-Nets, is impractical to manufacture, let alone at scale.

2/ Privacy is a complex cornerstone of society. Concerns from users, such as individuals that use medical robots, companies that use warehouse carts and governments that use drones, can prevent deployment of robots that would help people.

1/ Do you have a robot vacuum cleaner? Does it use a camera? Where do those images go? Is there a better way to keep this data private? In our paper, @dgdvision, @nikoSuenderhauf, Peter Corke and I sound a call to action for privacy-preserving vision.