{"id":95425,"title":"Parametric: RL for Robotics","tagline":"Robots that learn from customer feedback","body":"**TL;DR**\n\nExisting robots learn generic averages from massive datasets. Our robots learn from customer feedback.\n\nWe solve the RL feedback loop for robotics, allowing our robots to learn from _your_ specific standards via an automated reward pipeline that combines customer feedback with an intelligent judge model. This is the only way to capture the multi-trillion dollar “long tail” of unique, high-aggregate-value business tasks. \n\nWe’re live with real customers today and aggressively improving model performance. Our team led autonomy at Parallel Systems (series B startup; 100M raised) and scaled AI revenue for Facebook’s recommendation systems.\n\nhttps://youtu.be/EhiYJT52EJY\n\n**The Problem**\n\nThe real market for general purpose robotics is a multi-trillion dollar **“long tail” of differentiated tasks.** This is the market that traditional automation can’t touch. The core technology to solve this is _finally_ viable, but the long tail means the only path to success is adaptation to millions of specific customers' needs. \n\nWe’ve spoken to countless customers. No one has ever asked for a human form factor. They demand three things: **reliability, throughput, and control**. They want boxes packed in a specific order or towels folded lengthwise twice. Traditional learning algorithms don’t give this customer specificity.\n\nWhen you work backwards from these _actual_ customer requirements it becomes clear that we’re missing the bridge between customer feedback and model improvement.\n\n**Our Approach**\n\nWe build **robots** that learn from customer feedback and improve continuously. \n\n1. Adaptive Intelligence: At its core, an automated pipeline that combines customer feedback with a judge model to RL fine-tune a model and steer it towards positive behaviors, as defined by customers.\n2. Pragmatic Hardware: A mobile robot with a wheeled base and bimanual arms. Designed for long uptime and high throughput.\n\n \n\nOur models can learn new behaviors on-site with less than an hour of data, delivering low cost, effective automation immediately. \n\nIf you’d like to learn more, please reach out to us: [hello@parametric.company](mailto:hello@parametric.company).\n\n![uploaded image](/media/?type=post\u0026id=95425\u0026key=user_uploads/2757584/514106c9-df57-4f4e-aff8-de58700525d3)\n\n","slug":"Op7-parametric-rl-for-robotics","created_at":"2025-11-12T07:26:05.335Z","updated_at":"2026-04-25T17:25:52.432Z","total_vote_count":13,"url":"https://www.ycombinator.com/launches/Op7-parametric-rl-for-robotics","share_image_url":"https://www.ycombinator.com/media/?type=post\u0026id=95425\u0026key=user_uploads/2757584/514106c9-df57-4f4e-aff8-de58700525d3","company":{"id":30604,"name":"Parametric","slug":"parametric","url":"https://parametric.company/","logo":"https://bookface-images.s3.amazonaws.com/small_logos/6077089b2820cf087e262d7c3ff9da8b9617d791.png","batch":"Fall 2025","industry":"Industrials","tags":[],"search_path":"https://bookface.ycombinator.com/company/30604"}}