General Trajectory develops AI-powered software for industrial robots, enabling automation of tasks like palletizing, sorting, picking, and packing. Their solutions aim to improve efficiency and reduce labor costs in manufacturing and logistics operations.
Funding
$500K raised to dateRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Product
Problem
Many industrial robots require extensive manual programming and struggle to adapt to variations in object types, orientations, and environmental conditions, limiting their ability to automate complex tasks in dynamic environments. Traditional vision-language-action models often rely on expensive, human-collected data, capping performance and hindering generalization to new scenarios.
Solution
General Trajectory is developing general-purpose AI for industrial robots using vision-language-action (VLA) models trained with reinforcement learning from verifiable rewards. Their system leverages a "cold start" fine-tuning phase to familiarize a pre-trained vision-language model (VLM) with a chain-of-thought process, enabling the robot to reason through complex physical tasks. The AI policy generates its own reasoning and action traces during online rollouts in a simulated environment, using hardcoded verifiers to provide dense rewards for improved long-horizon planning. This approach enables robots to automate tasks such as mixed-SKU palletization, item picking, sorting, and packing, even with unseen objects.
Target Audience
The primary target audience includes companies in the manufacturing, logistics, and warehousing sectors seeking to automate complex physical tasks and reduce labor costs using AI-powered robots.
Features
- Vision-language-action (VLA) models that break down complex physical tasks using chain-of-thought reasoning
- Fine-tuned vision-language model (VLM) that outputs low-level robotic controls, including end-effector poses and gripper state represented as a 9D action vector
- Reinforcement learning from verifiable rewards using Nvidia's Isaac Sim for online rollouts and photorealistic data collection
- Ability to generalize to unseen tasks and objects due to pre-training on internet-scale data
- Models trained across different robot embodiments to automate valuable labor in logistics and manufacturing