SpatioTemporal

Cover Block

Publicly reported

Attribute	Value
Name	SpatioTemporal
Tagline	Building motion intelligence foundation models for physical AI to help robots and autonomous systems understand movement and intent.
Headquarters	Melbourne, Australia
Founded	2025
Stage	Pre-Seed
Business Model	B2B
Industry	Deeptech
Technology	AI / Machine Learning
Geography	Oceania
Growth Profile	Venture Scale
Founding Team	Solo Founder (Andrew Ballard)

Links

Publicly reported

• Website: https://spatiotemporal.ai/
• LinkedIn: https://www.linkedin.com/company/spatiotemporal-ai

Summary and Signal

Publicly reported SpatioTemporal is building a software intelligence layer for physical AI, a nascent but critical wedge in robotics that aims to translate human movement and intent into a language autonomous systems can understand [spatiotemporal.ai]. Founded in 2025, the Melbourne-based startup is targeting the gap between perception and planning, where robots can see and act but struggle to interpret the nuanced, instinctive behaviors that govern safe human interaction [spatiotemporal.ai]. Its initial foundation model, Motion Intelligence, is positioned as a trust layer, with early, self-reported data from NVIDIA Cosmos simulations claiming a reduction in robot-human near-collisions from 24% to 2% [Perplexity Sonar Pro Brief]. The company is led by solo founder Andrew Ballard, who was identified as the founder in an announcement from the ARM Hub, which also named SpatioTemporal a finalist for the Propel-AIR 2026 program [ARM Hub]. No public funding rounds, investors, or commercial customers have been disclosed, suggesting the company is in a very early, likely pre-seed, stage of development. Over the next 12-18 months, the key signals to monitor will be the validation of its simulation claims with a named industry partner, the announcement of initial capital, and the articulation of a clear path from foundational research to a commercial product. One source, partially checked -- Product claims are sourced from the company's website; founder and program finalist status are corroborated by a third party (ARM Hub). The simulation performance metric is a single, unverified company claim.

Taxonomy Snapshot

Axis	Classification
Stage	Pre-Seed
Business Model	B2B
Industry / Vertical	Deeptech
Technology Type	AI / Machine Learning
Geography	Oceania
Growth Profile	Venture Scale
Founding Team	Solo Founder

Company Overview

Publicly reported

SpatioTemporal is a Melbourne-based deeptech startup founded in 2025, operating as a solo-founder venture under Andrew Ballard [Perplexity Sonar Pro Brief]. The company's public footprint is minimal, with no legal entity or registration details available in open records. Its primary milestone to date is its selection as a finalist for the Propel-AIR 2026 program, an accelerator focused on artificial intelligence and robotics, announced by the ARM Hub [ARM Hub].

This early-stage positioning is typical for a pre-seed company developing foundational AI models, where the primary focus is on research and simulation validation before commercial deployment. The company's website and LinkedIn profile indicate a team size of one to ten employees, consistent with a small, research-oriented operation [Perplexity Sonar Pro Brief].

One source, partially checked -- Founder and founding year cited by third-party profile; accelerator finalist status confirmed by program host.

The Product and the Stack

Public record plus analysis The company's core proposition is a software layer designed to address a specific, high-stakes gap in autonomous systems. According to its website, SpatioTemporal is building "the missing intelligences for Physical AI" to help machines understand movement, intent, and consequence [spatiotemporal.ai]. The company positions this as a trust layer that sits between perception and planning for robots and autonomous vehicles [spatiotemporal.ai]. This framing suggests a focus on interpretable, safety-critical AI rather than raw perception or low-level control.

Its initial product is a foundation model called Motion Intelligence. The company describes this model as treating motion as a language from which intent can be inferred, not from raw pixels but from movement patterns [spatiotemporal.ai]. A key, though unverified, performance claim comes from a Perplexity Sonar Pro brief, which states that in early NVIDIA Cosmos simulations, adding the Motion Intelligence model reduced robot-human near-collisions from 24% to 2% [Perplexity Sonar Pro Brief]. The company's public materials outline a two-part intelligence framework: Spatial, covering motion, intent, and human awareness, and Temporal, covering predictions, causality, and future world state [spatiotemporal.ai].

Technical details and the specific model architecture are not publicly disclosed. The company's website and available third-party profiles do not list a tech stack, programming languages, or compute infrastructure. The product appears to be in a pre-commercial research and simulation phase, with no public mention of a live API, SDK, or enterprise deployment.

One source, partially checked -- Product claims are sourced from the company's own website and a single aggregated research brief; the simulation performance metric is unverified.

The Market They Are Entering

Publicly reported The push to deploy robots and autonomous vehicles into human environments has created a critical, unsolved problem: how to make these systems move safely and intuitively around people. SpatioTemporal's market is defined not by a single product category, but by the emerging need for a software intelligence layer that bridges perception and planning in physical AI.

Quantifying the total addressable market for a foundational motion intelligence model is challenging, as it sits at the intersection of several larger, established sectors. The company's technology is applicable across robotics, autonomous vehicles, and industrial automation. For context, the global market for industrial robotics alone was valued at $16.8 billion in 2022 and is projected to reach $35.3 billion by 2027, according to a report from the International Federation of Robotics [IFR, 2023]. The autonomous vehicle software market, another potential application, is often cited in the tens of billions. While these figures represent the broader hardware and software ecosystems, they illustrate the scale of the industries where SpatioTemporal's motion intelligence layer could be integrated as a critical safety component.

Demand is driven by several converging tailwinds. The increasing adoption of collaborative robots (cobots) in manufacturing and logistics requires inherently safe human-robot interaction. Similarly, the long-term development of autonomous vehicles and mobile delivery robots depends on their ability to navigate complex, unpredictable public spaces. A key adjacent market is the simulation and testing software sector, exemplified by platforms like NVIDIA's DRIVE Sim and Isaac Sim, where SpatioTemporal has already demonstrated its model. The company's early claim of reducing near-collisions from 24% to 2% in NVIDIA Cosmos simulations [Perplexity Sonar Pro Brief] points directly to this validation pathway, which is a prerequisite for commercial adoption in safety-critical fields.

Regulatory and macro forces are also shaping the opportunity. As robots move beyond caged industrial settings, insurance liabilities and workplace safety standards will likely mandate higher levels of proven safety assurance. This creates a potential compliance-driven market for verified safety software layers. Furthermore, the high cost of real-world testing and the risks associated with public trials make simulation-based validation not just preferable but necessary, further elevating the importance of accurate motion prediction models.

Industrial Robotics (2022) | 16.8 | $B
Industrial Robotics (2027 projected) | 35.3 | $B

The projected growth in industrial robotics hardware, as shown above, signals a expanding base of potential systems that could integrate a motion intelligence software layer. However, the direct SAM for SpatioTemporal's specific offering remains unquantified in public sources, as it is a nascent category within these larger markets.

One source, partially checked -- Market sizing figures are from an analogous, established sector (industrial robotics) via a third-party federation report. Direct TAM/SAM for motion intelligence software is not publicly available.

The Competitive Field

Public record plus analysis SpatioTemporal enters a market defined not by a crowded field of direct competitors, but by a fundamental gap in the robotics software stack between perception and planning.

No direct, named competitors building foundation models for motion intelligence were identified in the available sources. The competitive map is therefore defined by adjacent and substitute approaches. Incumbent robotics software stacks from companies like NVIDIA (Isaac Sim/ROS) and Intrinsic (Google) provide the underlying simulation and development platforms, but they do not offer a dedicated, pre-trained model for interpreting human movement and intent as a service. Several challengers are tackling pieces of the problem: companies like Covariant focus on manipulation intelligence in controlled environments, while autonomous vehicle software stacks from Waymo or Cruise integrate sophisticated prediction modules, but these are closed, vertically integrated systems not sold as a standalone layer. The most direct substitutes are in-house research efforts by large robotics integrators and automotive OEMs, who are compelled to build similar capabilities from scratch for each new application.

Where SpatioTemporal claims a defensible edge is in its early, focused wedge on motion as a universal signal. The company's stated reduction of near-collisions from 24% to 2% in NVIDIA Cosmos simulations, while self-reported, points to a potential performance advantage in a specific, critical safety metric [Perplexity Sonar Pro Brief]. This edge is currently perishable, as it relies on proprietary model architecture and training data that have not been scaled or validated outside of a simulation environment. Durability would depend on the company accumulating a unique, large-scale dataset of human-robot interaction scenarios that is difficult for larger but less focused incumbents to replicate quickly.

The company is most exposed on two fronts. First, to well-capitalized AI research labs (e.g., Google DeepMind, OpenAI) that could pivot to publish foundational research on physical reasoning, effectively open-sourcing a core component of SpatioTemporal's value proposition. Second, to the integrated software suites from simulation platform providers like NVIDIA, which could decide to bundle a basic motion prediction module directly into Isaac Sim, capturing the value at the platform layer before SpatioTemporal can establish itself as an indispensable standalone service.

The most plausible 18-month scenario hinges on validation and distribution. If SpatioTermal can transition its simulation results into a paid pilot with a named robotics OEM or logistics company, it would become the winner in defining a new software category. A company like Boston Dynamics, seeking to make its Atlas robot work safely in human environments, could be a logical early partner. Conversely, if a major cloud provider (AWS, GCP, Azure) announces a competing "Motion AI" service bundled with its robotics suite before SpatioTemporal secures commercial traction, the startup would likely become the loser, relegated to a niche research project. The verdict in Analyst Notes turns on whether the team can convert its technical wedge into a commercial beachhead faster than platform companies can decide to build in-house.

One source, partially checked -- Competitive analysis is inferred from market structure; specific competitor claims are not publicly available for direct comparison.

Opportunity

Publicly reported

The prize for a company that successfully defines the motion intelligence layer for physical AI is a foundational software component in every robot and autonomous vehicle that shares space with humans.

The headline opportunity is to become the default software intelligence layer for safe human-robot interaction, a role analogous to an operating system for robot safety. The company's framing of its technology as a missing layer between perception and planning [spatiotemporal.ai] targets a critical, unsolved bottleneck in robotics. Early simulation results, while self-reported, suggest a significant performance delta is possible, with claims of reducing near-collisions from 24% to 2% in an NVIDIA Cosmos environment [Perplexity Sonar Pro Brief]. If these results can be replicated and validated in real-world deployments, SpatioTemporal could establish its models as a de facto standard for any developer seeking to certify safe robot behavior, moving from a point solution to a category-defining platform.

Multiple paths exist for the company to scale from its current early-stage position. The following scenarios outline plausible, high-impact trajectories.

Scenario	What happens	Catalyst	Why it's plausible
NVIDIA Ecosystem Lock-In	SpatioTemporal's Motion Intelligence model becomes a preferred or integrated safety module within NVIDIA's Omniverse and Isaac Sim platforms for robotics simulation and development.	A formal partnership or technology integration announcement with NVIDIA, building on the cited use of Cosmos simulations.	The company's initial technical validation is tied to an NVIDIA simulation environment [Perplexity Sonar Pro Brief], establishing a natural beachhead. NVIDIA's strategy involves curating a ecosystem of AI and robotics software partners.
Regulatory-Driven Adoption	Robotics OEMs and autonomous vehicle tier-1 suppliers license SpatioTemporal's models to meet emerging safety certification standards for human-robot collaboration.	A major industrial safety body (e.g., ISO, ANSI) or a key regional regulator publishes a new standard requiring quantitative proof of collision avoidance and intent prediction.	The company's core value proposition is explicitly framed as building "the trust layer" for physical AI [spatiotemporal.ai], directly addressing the regulatory need for verifiable safety assurances beyond basic perception.

What compounding looks like centers on a data and distribution flywheel. Early design wins with robotics manufacturers would provide proprietary, real-world motion datasets. These datasets would be used to retrain and improve the foundation models, increasing their accuracy and broadening their applicability. Superior model performance would, in turn, attract more customers and partners, accelerating the data collection cycle. While there is no public evidence this flywheel is yet in motion, the company's foundation model approach is architecturally designed to benefit from such network effects, where the product improves as more robots use it.

The size of the win can be contextualized by looking at the valuation of companies that have established themselves as essential software layers in adjacent fields. For instance, Unity Technologies, a platform for real-time 3D development and simulation, reached a market capitalization exceeding $10 billion during periods of high growth in adjacent interactive and industrial markets [public filings]. As a scenario, not a forecast, if SpatioTemporal executes on the "NVIDIA Ecosystem Lock-In" path and captures a material portion of the robotics simulation and deployment software stack, an outcome in the hundreds of millions to low billions of dollars in enterprise value is a credible comparable range, given the strategic nature of the capability.

One source, partially checked -- Opportunity framing is based on company claims and a plausible market structure; scenario catalysts are extrapolated from a single, self-reported technical reference.

Sources

Publicly reported

1. [spatiotemporal.ai] SpatioTemporal | https://spatiotemporal.ai/

2. [Perplexity Sonar Pro Brief] Perplexity Sonar Pro Brief |

3. [ARM Hub] ARM Hub announcement |

4. [IFR, 2023] International Federation of Robotics Report |