This Robot Knows When to Back Off: Palm Garden AI Builds a Social Awareness Layer for Service Robots
A small startup says service robots already know how to move and speak. What they lack is the judgment to know when not to.
Key points
- Palm Garden AI, a company with offices in Germany and Thailand, has announced a software layer called Coherence Guard designed to help service robots read social situations.
- The system evaluates whether a robot's next action is socially appropriate, not just technically possible, before carrying it out.
- Coherence Guard runs directly on the robot where possible, so it doesn't need a cloud connection to make real-time judgment calls.
- The underlying framework, called the Relational Infrastructure Framework, draws on three years of human-interaction research at a retreat centre in Thailand.
- The software is currently available to hardware partners on request, with a commercial licensing model in preparation.
A hospital robot rolls toward a patient to deliver medication. The patient holds up a hand. A standard robot completes the delivery anyway, because the task says "deliver." Coherence Guard, a new software layer from Palm Garden AI, is designed to make the robot pause instead, read that signal, and back away.
The system sits on top of whatever software already controls a robot's movement and perception. It's a last-second sense-check. Before the robot acts, the layer evaluates whether the action is appropriate given who is nearby, how they're reacting, and what the social context demands.
CEO Joachim Scheuerer described it to The Robot Report as "an additional pre-action evaluation layer." The system reads signals including timing, proximity, emotional tone, and boundary requests. If those signals suggest the action would feel wrong, even if technically correct, the layer can instruct the robot to pause, explain itself, or step back.
Why does this matter for ordinary people?
Robots are entering spaces where misreading social cues causes real harm. Eldercare wards, hotel lobbies, and home environments all require a machine to understand not just what a person said but what they meant. A robot that ignores hesitation is at best annoying, at worst distressing.
Palm Garden AI developed the behaviour patterns behind Coherence Guard from three years of observation at Palm Garden Retreat in Thailand. Scheuerer's team studied arrivals, moments of silence, and trust-building. Respectful withdrawal, recognising discomfort and stepping back without fuss, emerged as a core pattern worth encoding.
Coherence Guard is built to run on ROS 2, the standard open software framework most professional robots already use. Manufacturers don't need to rebuild from scratch. The company's preferred deployment is local, on the robot itself, so a slow internet connection never delays a judgment call that matters in the moment. Cloud components handle simulation, logging, and fleet-level learning separately.
Coherence Guard doesn't replace the physical safety systems that stop a robot from crushing someone's hand. Scheuerer is clear that certified hardware safety stays separate. His layer handles a softer but equally real problem: knowing when a technically safe action is still the wrong thing to do. It's a distinction the Agility Robotics CEO also pressed in her six-point policy plan on 15 July, where social operating context was the gap nobody had yet closed with a product.
The core software is patent-pending and will not be fully open-source. Palm Garden AI is preparing a licensing model with optional cloud tools for testing and analysis. The Relational Infrastructure Framework is available to hardware partners now, on request.
What to watch: the behaviour patterns come from a retreat centre, not a hospital ward or a care home. Real validation will mean field deployments in exactly those messier settings.
