Intempus gives robots emotional intelligence so they can communicate failure states and intent to non-technical humans

Teddy, founder of Intempus, is building what he calls the interaction layer for industrial robots — the missing piece between a robot's internal failure states and the non-technical humans working around them.

The core problem is straightforward: industrial robots fail roughly 1% of the time, almost never for technically complex reasons. A warehouse robot gets blocked by a person, logs a fail code in the command line, and moves on. Nobody without engineering access knows what happened or why. Intempus sits on top of the robot's decision transformer as an augmentation layer, translating that internal state into expressive, human-readable signals — so a warehouse worker understands the robot's intent without reading a command line.

“Robots currently suck at communicating what they're doing. They're really hard to understand. You have to pay expensive nerds like me to use the command line anytime you're trying to understand what they're up to. What Intempus does is we allow these robots to be expressive — we allow them to convey their intent to people who would not otherwise understand what the robot is trying to do. One of the tricks behind making robots expressive is giving them emotions, because what are you expressing? You're expressing emotions — and you can view all of these as scalers.”

Go-to-market

Intempus targets enterprise robotics manufacturers — the companies selling robots to Amazon and similar operators — rather than the end operators themselves. Teddy's argument is that standardization matters here. If every robotics company ships its own interaction model, workers encounter three incompatible "languages" on the same factory floor, which compounds the usability problem rather than solving it. Selling through manufacturers is his path to proliferating a single standard.

Tech architecture

The technical mechanism borrows from how humans perceive time under stress. Intempus assigns robots what Teddy calls time constants — emotion-like scalars that govern compute allocation. When a robot is operating normally, it runs on minimum compute. When it enters a stress state — a failure, an edge case, an obstruction — it allocates more compute to collect richer data, surface the reason for failure, and communicate it clearly. The parallel to test-time inference scaling is deliberate: harder problems get more reasoning budget.

The emotional model extends further. Teddy describes giving robots adjustable parameters for joy, humor, and stress — scalars that shape how the robot presents itself. His reference point is TARS from Interstellar, which ships with a configurable humor level, simultaneously a cold ex-military machine and a trusted companion.

The longer ambition is consumer and service robotics, but for now Intempus is focused entirely on the industrial segment, where robots already exist in volume and the downtime costs of miscommunication are concrete and measurable.