DETAILED ACTION
This action is written in response to the remarks and amendments filed 1/29/26. This action is made final. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
The Applicants argue that the previous art of record does not anticipate or render obvious the claims as currently amended. The Examiner provides updated prior art rejections below necessitated by the current amendments.
Subject Matter Eligibility
In determining whether the claims are subject matter eligible, the examiner has considered and applied the 2019 USPTO Patent Eligibility Guidelines, as well as guidance in the MPEP chapter 2106. The examiner finds that the independent claims are directed to the practical application of compressing encoded vehicle data (eg for applications in a vehicle-to-vehicle communication network). Furthermore, the combination of steps performed in the recited method cannot be practically performed as a mental process.
Claim Rejections - 35 USC § 112(b) - Indefiniteness
The following is a quotation of the second paragraph of 35 U.S.C. 112:
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
Claims 1-20 are rejected under 35 U.S.C. 112(b), as being indefinite for failing to particularly point out and distinctly claim the subject matter which applicant regards as the invention.
Claim 1 recites, in part:
receive, by a subject vehicle, packets with compressed partial representations of a latent space associated with different views of target vehicles, wherein the compressed partial representations are internal to a first prediction model and encoded by the first prediction model;
However, this language introduces several ambiguous points:
If the partial representations received by the subject vehicle are already compressed, how are they internal to the first prediction model, and encoded by the first prediction model?
Is the first prediction model internal to the subject vehicle? External to the subject vehicle? Distributed across more than one location?
Because it is not clear which of the above interpretations is applicable, the term is ambiguous, and consequently a person of ordinary skill would not be able to understand the scope of the claim with reasonable certainty. Therefore the claim is indefinite. Independent claims 10 and 12 are rejected for the same reason, and all pending dependent claims inherit this deficiency.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103(a) which forms the basis for all obviousness rejections set forth in this Office action:
(a) A patent may not be obtained though the invention is not identically disclosed or described as set forth in section 102 of this title, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter pertains. Patentability shall not be negatived by the manner in which the invention was made.
The following are the references relied upon in the rejections below:
Bansal (Bansal, Ayoosh, Jayati Singh, Micaela Verucchi, Marco Caccamo, and Lui Sha. "Risk ranked recall: Collision safety metric for object detection systems in autonomous vehicles." In 2021 10th Mediterranean conference on embedded computing (MECO), pp. 1-4. IEEE, 2021.)
Liu (US 9,178,593 B1)
Wang (Wang, Tsun-Hsuan, Sivabalan Manivasagam, Ming Liang, Bin Yang, Wenyuan Zeng, and Raquel Urtasun. "V2vnet: Vehicle-to-vehicle communication for joint perception and prediction." In European conference on computer vision, pp. 605-621. Cham: Springer International Publishing, 2020.)
Xu (Xu, Runsheng, Hao Xiang, Xin Xia, Xu Han, Jinlong Li, and Jiaqi Ma. "Opv2v: An open benchmark dataset and fusion pipeline for perception with vehicle-to-vehicle communication." In 2022 International Conference on Robotics and Automation (ICRA), pp. 2583-2589. IEEE. 23-27 May 2022.)
Claims 1-5, 7, 9-16, 18 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Xu and Wang.
Regarding claims 1, 10 and 12, Xu discloses an assistance system (and a related computer-readable medium and method) comprising:
a processor; and a memory storing instructions that, when executed by the processor, cause the processor to: …
A processor and a memory are inherent throughout the Xu disclosure.
generate, by an attention network, an attention vector about the different views by aggregating the packets for the target vehicles; …
P. 2587, first col. “Intermediate fusion: The Attentive Fusion pipeline is flexible and can be easily generalized to other object detection networks. To evaluate the proposed pipeline, we only need to add the Compression, Sharing, and Attention (CSA) module to the existing network architecture. Since 4 different detectors add CSA modules in a similar way, here we only show the architecture of intermediate fusion with the PIXOR model as Fig. 6 displays. Three CSA modules are added at the 2D backbone of PIXOR to aggregate multi-scale features while all other parts of the network remain the same.” (Emphasis added.)
train the second prediction model using the addition vector to reduce data representations and adapt data compression associated with the latent space.
P. 2586, “The overall architecture is shown in Fig. 5. The proposed pipeline is flexible and can be easily integrated with existing Deep Learning-based LiDAR detectors (see Table III).”
See also caption to fig. 5: “5) Attentive Fusion: leverage self-attention to learn interactions among features in the same spatial location. 6) Prediction Header: generate final object predictions.”
Wang discloses the following further limitation which Xu does not disclose:
receive, by a subject vehicle, packets with compressed partial representations of a latent space associated with different views of target vehicles, wherein the compressed partial representations are internal to a first prediction model and encoded by the first prediction model; …
P. 607, fig. 2, (reproduced below).
PNG
media_image1.png
276
818
media_image1.png
Greyscale
“receive… packets” :: “Receive V2V Messages”. The Examiner notes that this information is compressed, as shown by the ‘Compressor’ box.
compute, by a second prediction model, an addition vector that optimizes data decoding by the second prediction model using acquired data from the attention vector; and
P. 610, algorithm 1. The Examiner interprets “addition vector” according to its broadest reasonable interpretation in view of the specification. The Applicant provides no definition for this term. Accordingly, the Examiner interprets this term as encompassing the learned hyperprior discussed in the passage below:
P. 610, “Compression: We now describe how each vehicle compresses its intermediate representations prior to transmission. We adapt Balle et al.’s variational image compression algorithm [2] to compress our intermediate representations; a convolutional network learns to compress our representations with the help of a learned hyperprior. The latent representation is then quantized and encoded losslessly with very few bits via entropy encoding. Note that our compression module is differentiable and therefore trainable, allowing our approach to learn how to preserve the feature map information while minimizing bandwidth.” (Emphasis added.)
P. 613, “We use a rate-distortion objective, which aims to maximize the bit rate in transmission while minimizing the distortion between uncompressed and decompressed data. We define the rate objective as the entropy of the transmitted code, and the distortion objective as the reconstruction loss (between the decompressed and uncompressed feature maps).” (Emphasis added.)
At the time of filing, it would have been obvious to a person of ordinary skill to combine the V2V message compression technique disclosed by Wang with the V2V communication system of Xu because compression can reduce the bandwidth required for communications, which could result in more reliable communications, less computing resources used and ultimately safer roads.
Regarding claims 2, 11 and 13, Xu discloses the further limitation wherein the instructions to train the prediction model further include instructions to adapt the prediction model … for feature decoding associated with the target vehicles during an intersection crossing.
P. 2584, fig. 1, illustrating an intersection scenario for an AV.
PNG
media_image2.png
410
888
media_image2.png
Greyscale
Wang discloses the following further limitation which Xu does not disclose:
… concatenation of the addition vector to processed data …
P. 610, “Compression: We now describe how each vehicle compresses its intermediate representations prior to transmission. We adapt Balle et al.’s variational image compression algorithm [2] to compress our intermediate representations; a convolutional network learns to compress our representations with the help of a learned hyperprior. The latent representation is then quantized and encoded losslessly with very few bits via entropy encoding. Note that our compression module is differentiable and therefore trainable, allowing our approach to learn how to preserve the feature map information while minimizing bandwidth.”
P. 613, “We use a rate-distortion objective, which aims to maximize the bit rate in transmission while minimizing the distortion between uncompressed and decompressed data. We define the rate objective as the entropy of the transmitted code, and the distortion objective as the reconstruction loss (between the decompressed and uncompressed feature maps).” (Emphasis added.)
Regarding claims 3 and 14, Xu discloses the further limitation including instructions to aggregate, by the subject vehicle, the packets by splitting the latent space using a quality factor associated with the different views of one of the target vehicles.
P. 2587, fig. 5 (reproduced below): “Fig. 5: The architecture of Attentive Intermediate Fusion pipeline. Our model consists of 6 parts: 1) Metadata Sharing: build connection graph and broadcast locations among neighboring CAVs. 2) Feature Extraction: extract features based on each detector’s backbone. 3) Compression (optional): use Encoder-Decoder to compress/decompress features. 4) Feature sharing: share (compressed) features with connected vehicles. 5) Attentive Fusion: leverage self-attention to learn interactions among features in the same spatial location. 6) Prediction Header: generate final object predictions.”
PNG
media_image3.png
388
936
media_image3.png
Greyscale
Regarding claims 4 and 15, Xu discloses the further limitation wherein the instructions to train the prediction model further include instructions to reduce dimensionality and losses of the data compression using the addition vector.
P. 2586, second col. “Compression and Feature sharing: An essential factor in V2V communication is the hardware restriction on transmission bandwidth. The transmission of the original high-dimensional feature maps usually requires large bandwidth and hence compression is necessary. One key advantage of intermediate fusion over sharing raw point clouds is the marginal accuracy loss after compression [15]. Here we deploy an Encoder-Decoder architecture to compress the shared message. The Encoder is composed of a series of 2D convolutions and max pooling, and the feature maps in the bottleneck will broadcast to the ego vehicle.”
Regarding claims 5 and 16, Xu discloses the further limitation wherein the addition vector has an intermediate representation reflected by samples from the different views and the different views form a statistical distribution.
‘different views’ :: P. 2588, second col. “E. Effect of CAV [connected automated vehicles] Quantity We explore the detection performance as affected by the number of CAVs in a complex intersection scenario where 150 vehicles are spawned in the surrounding area. A portion of them will be transformed into CAVs that can share information. We gradually increase the number of the CAVs up to 7 and apply VoxelNet with different fusion methods for object detection. As shown in Fig. 7, the AP has a positive correlation with the number of CAVs. However, when the quantity reaches 4, the increasing rate becomes lower. This can be due to the fact that the CAVs are distributed on different sides of the intersection and four of them can already provide enough viewpoints to cover most of the blind spots. Additional enhancements with 5 or more vehicles come from denser measurements on the same object.”
P. 2585, second col. “Fig. 3 and Fig. 4 reveal the statistics of the 3D bounding box annotations in our dataset. Generally, the cars around the ego vehicle are well-distributed with divergent orientations and bounding box sizes. This distribution is in agreement with the data collection process where the object positions are randomly selected around CAVs and vehicle models are also arbitrarily chosen.”
Regarding claims 7 and 18, Xu discloses the further limitation including instructions to adapt the data compression according to the prediction model detecting one of an object and trajectories associated with the target vehicles.
P. 2584, first col. “To meet requirements of both bandwidth and detection accuracy, intermediate fusion [22], [15] has been investigated, where intermediate features are shared among connected vehicles and fused to infer the surrounding objects. F-Cooper [22] utilizes max pooling to aggregate shared Voxel features, and V2VNet [15] jointly reason the bounding boxes and trajectories based on shared messages.”See also fig. 1 (reproduced supra).
Regarding claims 9 and 20, Xu discloses the further limitation wherein the packets are intermediate data encoded by the prediction model and include trajectory data associated with the target vehicles.
P. 2584, “F-Cooper [22] utilizes max pooling to aggregate shared Voxel features, and V2VNet [15] jointly reason the bounding boxes and trajectories based on shared messages.” (Emphasis added.)
Claims 6 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Xu, Wang and Liu.
Regarding claims 6 and 17, Liu discloses the further limitation which Xu/Wang do not disclose wherein the addition vector has an intermediate representation reflected by a mean and a variance from a Gaussian model of the acquired data.
Col. 12, line 29 et seq. “However, rather than reporting measurements for all directions as in the measurement report IE 230, a bitmask 252 included in the Directional Measurement field indicates a report of an average among directions metric, a variance among directions metric, and a strongest direction measurement (in FIG. 6, example values of flags are listed in italics, and length in octets is listed in normal font below the corresponding field or sub-field).”
At the time of filing, it would have been obvious to a person of ordinary skill to apply the Gaussian distribution-based bitmask technique disclosed by Liu to the combined system of Xu/Wang because this could result in better data compression rate, thereby reducing bandwidth requirements for effective communication.
Claims 8 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Xu, Wang and Bansal.
Regarding claims 8 and 19, Bansal discloses the further limitation which Xu/Wang do not disclose including instructions to rank, by nearby vehicles, samples from the addition vector within a hierarchy according to the target vehicles approaching an intersection including the nearby vehicles.
P. 1, second col. “This work introduces the R3 metrics, a risk aware version of Recall. The risk rankings are based on the risk of collision (§IV). Recall is measured for each rank separately. R3 1: Recall for objects that pose an imminent risk of collision (§IV-A). R3 2: Recall for objects that can potentially collide with the ego vehicle (§IV-B). R3 3: All other objects in the environment.”
At the time of filing, it would have been obvious to a person of ordinary skill to apply the technique disclosed by Bansal for ranking perceived risks (in an automated driving system) with the combined system of Xu/Wang because this could result in few collisions.
Additional Relevant Prior Art
The following references were identified by the Examiner as being relevant to the disclosed invention, but are not relied upon in any particular prior art rejection:
Ko discloses (US 2021/0133570) discloses a bitmask compression technique (see [0174]) with applications to autonomous vehicles (see [0085]).
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Vincent Gonzales whose telephone number is (571) 270-3837. The examiner can normally be reached on Monday-Friday 7 a.m. to 4 p.m. MT. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Miranda Huang, can be reached at (571) 270-7092.
Information regarding the status of an application may be obtained from the USPTO Patent Center.
/Vincent Gonzales/Primary Examiner, Art Unit 2124
Prosecution Insights