A robot can see a person. It can plan a path. But it cannot read a hesitation or anticipate a sudden turn. That gap, known as Moravec's Paradox, is where SpatioTemporal is placing its bet. The Melbourne-based startup is building what it calls the missing intelligence layer for physical AI, a foundation model designed to turn movement into machine-readable patterns that predict intent [spatiotemporal.ai].
Its early claim is stark. In NVIDIA Cosmos simulations, adding its Motion Intelligence model reportedly reduced robot-human near-collisions from 24% to 2% [Perplexity Sonar Pro Brief]. For a field where safety is the primary barrier to adoption, a 22-point swing is a number that gets attention.
The Missing Layer
SpatioTemporal frames its technology as a software wedge between a robot's perception and planning systems. While cameras and lidar tell a machine what is there, and algorithms chart a course, the company argues there is a critical blind spot: understanding the fluid, social dynamics of human movement. Its models aim to interpret cues like gait, speed changes, and relative positioning to infer whether a person is about to stop, change direction, or yield [spatiotemporal.ai].
An Early Validation Signal
Public traction is limited, as is typical for a 2025-founded deeptech company. The simulation results are self-reported and lack independent verification. There are no announced customers or commercial partnerships. The founder, Andrew Ballard, has a thin public record, with no prior company history or detailed bio available in the sources.
Yet the company has secured a notable early signal. It was selected as a finalist for the Propel-AIR 2026 program run by ARM Hub, an Australian advanced robotics consortium [ARM Hub]. Such programs often provide non-dilutive grant funding, technical resources, and industry connections. For a solo founder, this kind of institutional backing can be a crucial accelerant, helping to bridge the gap from simulation to a deployable prototype.
The Path to Proof
For now, the company appears to be in a pre-seed or bootstrapped phase. No funding rounds, investors, or valuations are disclosed. The path forward hinges on converting its simulation promise into a hardened product and securing its first paid pilot. The Propel-AIR selection suggests it is on the radar of relevant gatekeepers.
The next 12 months will test whether a focused startup can out-innovate well-funded internal teams at larger companies. Can SpatioTemporal translate a 2% collision rate in a digital environment into a tangible safety improvement on a real factory floor? The answer will determine if this is a research project or the beginning of a new layer in the autonomy stack.