The Provisional That Said Too Little
Your client, Kauri Robotics Pty Ltd (Australian), filed a provisional application on 3 March 2023 describing a warehouse picking robot with a suction gripper and a vision system that identifies box edges. The provisional contains one paragraph on 'machine learning to improve grasp selection' with no architecture, training data or algorithm. On 2 March 2024 Kauri files a complete application and a PCT application claiming priority from the provisional. The complete specification adds a convolutional neural network trained on 40,000 labelled images, a specific loss function, and a fallback claim to the suction gripper alone. In January 2025 a competitor publishes a paper (submitted 10 February 2023, published online 1 June 2023) disclosing a suction gripper robot with edge-detection vision. Kauri's examiner in Australia has issued a first report citing the paper as novelty-destroying prior art against all claims. Kauri also wants to validate in Germany and the US.
1. Advise whether the priority claim from the provisional is valid for (a) the neural-network claims and (b) the suction-gripper claim. What is the consequence for each?
Read the worked answer
The priority claim is assessed claim-by-claim and requires that the earlier application provide support for the claimed subject matter in the sense of the same invention being disclosed, not merely a generic statement of aim (see the support requirement applied in Australian practice under s 43(2) and the Pfizer / Lockwood line of authority, and the corresponding 'same invention' test in the UK and EPO).
(a) Neural-network claims. The provisional's single sentence about 'machine learning to improve grasp selection' does not disclose the CNN, the training set, or the loss function. A skilled reader could not derive the claimed invention from it. Priority is not valid; the effective date for these claims is 2 March 2024.
(b) Suction-gripper claim. If the provisional discloses the gripper with sufficient detail (it apparently does, at least as an apparatus), priority is valid and the effective date is 3 March 2023.
Consequences. The competitor paper was submitted 10 February 2023 but published 1 June 2023. For the gripper claim (priority 3 March 2023), the paper is not part of the prior art base because it was not published before the priority date. For the neural-network claims (effective date 2 March 2024), the paper published 1 June 2023 is prior art and anticipates or renders obvious those claims unless distinguishing integers (the CNN, loss function) confer novelty and inventive step. The examiner's report is therefore over-broad: the gripper claim should be defended on priority, and the neural-network claims must be narrowed or argued on the distinguishing features. Note that the added matter rules (s 102 / EPC Art 123(2)) also constrain amendment, since the CNN detail was not in the provisional.
2. What is the practical significance of the PCT filing date and the 12-month window, and what would you have advised in February 2024?
Read the worked answer
The PCT application filed 2 March 2024 is within the 12-month Paris Convention priority period from 3 March 2023, so it can validly claim that priority (Paris Art 4(C)(1); PCT Art 8). Filing the complete and the PCT on the same day is efficient but does not cure the priority defect for the neural-network claims: the PCT application inherits the same priority document and the same support problem.
Advice that should have been given in February 2024:
- File a second provisional or an early complete capturing the CNN architecture, training data and loss function before 3 March 2024, and claim priority from that for the neural-network claims. That would have given those claims a date before the 1 June 2023 publication - impossible, since the publication already occurred. In fact no filing in 2024 can beat a June 2023 publication.
- The real lesson is earlier: the machine-learning content should have been captured in the provisional or in a supplementary filing before 1 June 2023. By February 2024 the paper was already citable against any claim not entitled to the March 2023 date.
- Consider the grace period under s 24 (and equivalent provisions abroad): it is narrow (e.g. publication derived from the applicant, or at a recognised exhibition) and does not cover an independent third-party paper.
- Foreign strategy: Germany/EPO applies the same priority-support analysis; the US allows continuation and broader priority practice but still requires written description support (35 USC 112(a)). The US may allow a CIP-style approach, but the new matter gets the later date.
- Commercial step: consider narrowing to the gripper plus the specific CNN implementation and arguing inventive step over the paper on the loss function and training regime.