About
Instance builds automated evaluation software for robot policies: given a task description and video rollouts, it returns success verdicts with evidence and detailed subtask captions. It targets robotics teams training policies, running evaluations, or deploying robots, differentiating through automated, human-free verification across robots, cameras, and episodes.
Market
Instance competes in robotics data infrastructure and robot-learning evaluation, positioning itself as a verification layer that converts raw robot rollouts into success verdicts and trusted training data. Its differentiation is a specialized fine-tuned model for robot-task verification that can run locally on one GPU, provide evidence and grounded subtasks, and work across robots and camera angles; the listed competitors are broader adjacent alternatives in model deployment, robotics data annotation, and robot-foundation-model data pipelines.
Instance is aimed at robotics R&D teams, robot-learning labs, and companies generating real-world robot episodes that need automated task-success labels and trusted training data. The likely buyers are robotics or machine-learning research and engineering leads responsible for evaluating rollouts and improving robot policies.
At a Glance
Problem
Robot-learning teams are bottlenecked by the slow, repetitive process of evaluating whether a robot actually completed a task. A typical rollout requires a robot to attempt a task, a person to inspect the result and judge success, and often another reset before the next attempt. This human-in-the-loop workflow consumes scarce engineering and operations time, slows the collection of trustworthy training data, and makes policy iteration expensive. The clearest use case is evaluating large numbers of robot-policy rollouts during training or before real-world deployment, especially for tasks such as folding, stacking, picking, or placing objects.
Product / Service
Instance is building an automated verification layer for robot learning. Given a task description and camera footage from a rollout, its success detector returns a success verdict with evidence and detailed subtask captions, supporting any robot and camera angle. The company presents this as a software/API workflow that turns raw rollouts into trusted, labeled data, allowing teams to evaluate policies faster and reduce the need for manual judgment; its longer-term vision is an autonomous evaluation rig in which one robot performs the task, Instance judges the result, and another robot resets the scene.
Market
Instance competes in the emerging robotics infrastructure market, specifically automated robot-policy evaluation, rollout verification, and data quality for robot learning. Its closest alternatives are likely internal human review and adjacent evaluation or simulation platforms rather than a clearly established identical incumbent. NVIDIA’s RoboLab is pursuing diagnostic evaluation of real-world robot policies using simulation, while Runway describes using general world models for robot-policy evaluation and training, making those examples of adjacent competitive approaches rather than confirmed direct competitors.
The company is an early-stage venture-backed startup in YC’s Summer 2026 batch, with an active status, two employees, and zero listed jobs. It has launched the first step of its product—a success detector—and offers a live demo and founder-led demonstrations, but the available public materials disclose no customers, revenue, funding amount, or production deployment. On the evidence available as of August 1, 2026, Instance is best characterized as pre-commercial or at the initial customer-validation stage, with traction demonstrated by its public launch and working demo rather than reported commercial metrics.
Founders & Leadership
Funding History
Y Combinator
Recent News
Y Combinator partner Ankit Gupta announced Instance’s launch and showcased its success detector for robot policies. The product lets users describe a task and upload or record a video so the system can judge whether the robot completed it.
Instance launched its first product: a success detector that automatically determines whether a robot completed its task, with detailed subtask breakdowns. The company is building toward fully automated robot evaluation, including rollout assessment and scene resets without humans in the loop.
Active Roles
0No active roles right now.
Get notified when they postBusiness Model
Instance appears to use a demo-led B2B software model, selling its automated robot-policy verification and evaluation product to robotics teams. The company directs prospective users to book founder-led demos; no public pricing or specific subscription/API terms were disclosed in the gathered sources.