Nobody Has Left a Robot Working Overnight | Yulia Sandamirskaya, ZHAW
Prof. Dr. Yulia Sandamirskaya of ZHAW on neuromorphic computing, why AGI has been almost here since 1956, the case against humanoid robots and bringing care robots into Swiss elderly care.
Discuss this episode and meet other Swiss tech builders in the TechDrive WhatsApp group.
Key insights: executive summary
- Cleaning and greeter robots are real products, but no company has yet dared to leave a robot with a manipulator working overnight in a care facility. That is the line between a demo and a product.
- Intelligence is closer to imagination bandwidth than to data volume. A fruit fly performs acrobatics on roughly 100,000 neurons, and the stable picture in front of your eyes is about 95% imagination.
- AGI has been described as almost here since the Dartmouth summer school of 1956. Bigger static models alone are unlikely to close the remaining gap.
- Humanoid robots on legs are a hazard around people. The practical form factor is wheels plus arms with modular actuators, and a software layer like Neuraverse where robot skills work like apps.
- The breakthrough in elderly care robotics came from process, not technology: her team toured a care home, asked the staff why they needed robots, and a caregiving team leader started listing tasks until she had to get her notebook out.
“Nobody in the world has yet dared to leave a robot with a manipulator doing real work in a care facility and then gone to bed.”
“AGI has been almost there since 1956, and I do not believe bigger static models get you the rest of the way.”
Why has nobody left a robot working overnight in a care home?
Cleaning robots and greeter robots are commercially real, but anything with an arm is not yet a product you can buy and trust. The distance between a successful demo and an unattended overnight shift is the real frontier of physical AI, and crossing it requires reliability, safety evidence and trust from the people on site, not just better algorithms.
Her team's answer was to start with the caregivers rather than the robot. A tour of a care home ended in the staff room, where a team leader listed task after task he would hand to a machine. That list, captured on the spot, is the product roadmap.
Ask the staff first
Involve the people on site from the beginning, because their task lists reveal where robots genuinely help and where they would only add friction.
Design for acceptance
How a robot approaches, moves and signals intent determines whether care workers and residents accept it as a colleague or reject it as a disturbance.
What can neuromorphic computing teach us about real intelligence?
Neuromorphic systems take their cues from brains rather than data centres. A fruit fly does acrobatics on about 100,000 neurons, event-based cameras inspired by Misha Mahowald's work transmit only change instead of frames, and the human fovea covers only two degrees of your visual field. Most of what you perceive as a stable picture is reconstructed, which suggests intelligence is imagination bandwidth more than data volume.
This is a different target from the one the industry is chasing. Scaling static models adds trillions of connections, yet misses the sparse, time-based processing that lets biological systems act in the real world on a few watts.
Why is she betting on wheels, arms and robot apps instead of humanoids?
A legged humanoid on stairs is a hazard to everyone around it. Her pick for care and service work is a stable wheeled base with arms and modular actuators, paired with a software ecosystem like Neuraverse where robot skills are installed and shared like apps. The business model points towards robot as a service, which removes the prohibitive upfront cost for care homes.
Who is Yulia Sandamirskaya?
Prof. Dr. Yulia Sandamirskaya is a professor at ZHAW, the Zurich University of Applied Sciences, where she heads the Centre for Cognitive Computing in Life Sciences. Trained as a physicist in Bochum, she worked on Intel's Loihi neuromorphic chip before choosing the professorship in Zurich, and she is now spinning a company out of her lab to bring robots into elderly care.
Episode show notes
Yulia Sandamirskaya makes the case that physical AI's hardest problems are biological in inspiration and social in practice. We get into neuromorphic computing, event-based vision, multimodality and self-calibration, and why the stable picture in front of your eyes is mostly imagination.
We also cover what is genuinely deployed in care homes today, why she argues against humanoids on legs, the Neuraverse vision of robot skills as apps, robot-as-a-service pricing, and the personal motivation behind the work: her grandmother in Belarus, and the fifteen years it took the industry to build a vacuum cleaner that actually works.
More conversations from the Swiss tech ecosystem
- Physical AI Is Ready, Buildings Are Not | Kateryna Portmann, ANYboticsKateryna Portmann, ANYbotics
- Switzerland's AI Movement Is About to Start | Swiss {ai} WeeksKatka Letzing and Anna Hakobyan, Swiss {ai} Weeks
- Five Years to Teach a Robot Table Tennis | Mireille El Geche, Sony AIMireille El Geche, Sony AI
Follow the Swiss tech ecosystem, one drive at a time
New in-car interviews with founders, investors and researchers building Zurich's deep tech, AI and robotics scene. Want to discuss this episode, ask follow-up questions and meet other builders?