General Trajectory

What do they actually do

General Trajectory is training multimodal “vision‑language‑action” models that output a short natural‑language reasoning trace and an action sequence for robotic manipulation. The models are trained and evaluated in simulation (NVIDIA Isaac Sim). Today, the company is at an early‑prototype stage with a public waitlist and early pilot interest/LOIs, rather than a broadly shipped product or platform company site YC listing YC/LOIs tweet.

Their workflow starts from a pre‑trained vision‑language model and fine‑tunes it to emit alternating reasoning and action tokens for a task. They run online rollouts in simulation, use scripted/verifier checks to score intermediate steps, and iterate training to improve long‑horizon planning and generalization across objects, scenes, and robot embodiments before broader real‑world deployment company site.

In the near term they are targeting industrial logistics tasks—mixed‑SKU palletization, sorting, picking, and packing—working toward pilots with factories/warehouses using mobile manipulators and other robot forms company site YC listing YC LinkedIn post.

Who are their target customer(s)

• Warehouse operations manager at an e‑commerce or 3PL facility: Needs robots that keep working as SKUs, layouts, and packing patterns change. Current automation breaks under variation, and temp labor is costly and disruptive.
• Robotics/automation engineer running robot pilots: Spends significant time hand‑coding or re‑training task‑specific controllers. Multi‑step tasks don’t transfer reliably from lab to floor, and support for different grippers/bases is brittle.
• Head of automation procurement at a logistics company: Faces long, risky pilots with unclear ROI. Requires easy integration with existing fleets/WMS and clear validation artifacts before approving rollouts.
• System integrator or robotics integrator: Delivers bespoke solutions per site with high engineering hours and limited software reuse. Wants a reusable model to reduce per‑deployment effort.
• Quality/safety manager at an automated warehouse: Needs predictable, verifiable behavior and understandable failure modes for certification. Current black‑box controllers are hard to debug across long pick/place sequences.

How would they acquire their first 10, 50, and 100 customers

• First 10: Convert waitlist and YC introductions/LOIs into paid, co‑developed pilots on partner robots with one clear KPI (e.g., mixed‑SKU palletization throughput). Use simulation pre‑validation and a short acceptance test to define success company site YC/LOIs tweet.
• First 50: Offer a fixed, low‑risk pilot package (timeline, deliverables, milestone‑tied pricing) and sell through select integrators plus direct outreach to 3PL/e‑commerce ops. Publish a few reference case studies and a technical checklist to cut per‑site engineering company site YC LinkedIn post.
• First 100: Productize into a subscription with model updates, an integration SDK, and a certified partner network (OEMs + integrators). Provide an ROI calculator and standardized safety/verification artifacts to streamline procurement approvals company site YC listing.

What is the rough total addressable market

Top-down context:

Analysts estimate the global warehouse automation market at about USD 21.7B in 2024, with picking/packing functions representing a large share (around 29%) of spend Allied Mordor.

Bottom-up calculation:

Applying ~29% to USD 21.7B implies roughly USD 6.3B for pick/sort/palletize functions in 2024. Assuming software+integration is ~30–40% of that yields an immediately addressable TAM near USD 1.9–2.6B Allied Mordor Fortune Business Insights.

Assumptions:

• 2024 global warehouse automation baseline ≈ USD 21.7B (within common ~USD 20–26B range).
• Pick/sort/palletize share ≈ 29% of automation spend; GT targets this function slice.
• Software + integration accounts for ~30–40% of function spend relevant to GT’s offering.

Who are some of their notable competitors

• Covariant: Trains large robotics models (“Covariant Brain” / RFM‑1) for warehouse tasks like picking and depalletizing; has moved into commercial deployments and targets the same logistics buyers site customer stories.
• Berkshire Grey: Provides turnkey AI‑enabled automation systems (pick cells, sortation, palletizing, mobile fleets) to large retailers/3PLs—competes on end‑to‑end outcomes and production rollouts solutions customers.
• RightHand Robotics: Specializes in production piece‑picking (RightPick) with grippers, vision, and fleet software; strong in core picking/sorting deployments products resources.
• Plus One Robotics: Offers AI perception and supervised autonomy (PickOne/Yonder) for induction, depalletizing, and palletizing with various robot arms and integrators site PickOne.
• Intrinsic (Alphabet): Developer platform and AI capabilities (e.g., Flowstate) to make industrial robots easier to program and adapt across tasks and robot types—competes at the software platform layer site Flowstate.