Problem

Waddle Labs addresses the deployment bottleneck in robot learning. The prevailing approach relies on huge robot datasets and end-to-end models that are difficult to steer and do not reliably generalize across robots, environments, and tasks. As a result, adapting a robot to a new workspace or task typically requires fresh data collection and fine-tuning, creating substantial engineering time and deployment cost, although Waddle has not published a dollar estimate for that burden. The killer use case is quickly creating manipulation policies for real hardware—for example, placing microswitches into slots or packaging a box—rather than spending weeks building and tuning task-specific robot software.

Product / Service

Waddle provides an agent-based robotics-control API. A user connects a robot, describes a task in natural language, and the agents decompose the goal into subtasks, inspect camera feeds, write control code, and call specialist action models such as vision-language-action models. The output is an executable program that can be run, tested, and iterated through further conversation, with the company’s YC profile claiming a working policy can be produced in roughly 20 minutes.

The product is intended to combine the generalization of foundation models with the speed and reliability of programmed control. Waddle’s agents continuously create and refine reusable skills—such as grasping, aligning, and folding—so each solved task can expand a shared skill library. Beyond policy creation, the company highlights autonomous data generation and overnight robotics experimentation as important uses, allowing robots to run repeated trials and train or improve policies with less hands-on intervention.

Market

Waddle competes in agentic robotics-control software and the broader physical-AI or robot-foundation-model market. Its approach is differentiated from a model that maps an instruction directly to motor commands: Waddle uses code as the policy while retaining the ability to call action models as tools. The nearest visible alternatives are adjacent robotics-intelligence platforms such as Intrinsic, which describes itself as a software and AI robotics company building intelligent-automation infrastructure, and Physical Intelligence, whose π models are general-purpose or steerable robotic foundation models. The available sources do not identify a named direct competitor with an identical agent-and-API approach.

The company is at an early commercialization stage. Waddle Labs was founded in 2025 by Yiding Song and Hanming Ye, is listed by Y Combinator in its Summer 2026 batch as an active Boston company with two employees, and invites users to try a preview API. Its publicly described traction consists of internal demonstrations—creating policies, generating training data, and facilitating auto-research—rather than disclosed customer, revenue, or scale metrics, so it is best characterized as pre-revenue or at least pre-scale based on the available evidence.