Multi-modal 3-D pose estimation
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for estimating a 3-D pose of an object of interest from image and point cloud data. In one aspect, a method includes obtaining an image of an environment; obtaining a point cloud of a three-dimensional region of the environment; generating a fused representation of the image and the point cloud; and processing the fused representation using a pose estimation neural network and in accordance with current values of a plurality of pose estimation network parameters to generate a pose estimation network output that specifies, for each of multiple keypoints, a respective estimated position in the three-dimensional region of the environment.
1. A method performed by one or more computers, wherein the method comprises:
generating a fused representation of an image of an environment and a point cloud of a three-dimensional region of the environment, wherein the point cloud comprises a plurality of data points, and wherein the generating comprises:
generating, based on the image and for each of a plurality of keypoints, a score for each of a plurality of locations in the image that represents a likelihood that the keypoint is located at the location in the image; and
generating, for each of the plurality of data points in the point cloud, a respective feature vector that includes at least some of the scores generated based on the image;
processing the fused representation to generate, for each of the plurality of keypoints, an estimated position in the three-dimensional region of the environment; and
controlling an agent based at least on the estimated position for each of the plurality of keypoints.
2. The method of claim 1 , wherein controlling the agent based at least on the estimated position for each of the plurality of keypoints comprises:
generating, based at least on the estimated position for each of the plurality of keypoints, a planning decision for the agent; and
controlling the agent to implement the planning decision by transmitting one or more electronic signals that have been generated in accordance with the planning decision to one or more control units of the agent.
3. The method of claim 1 , wherein the agent comprises a vehicle, and wherein the environment is an environment in a vicinity of the vehicle.
4. The method of claim 1 , wherein the plurality of keypoints collectively define an estimated pose of each of one or more pedestrians in the environment.
5. The method of claim 3 , wherein the vehicle comprises an autonomous vehicle, or a semi-autonomous vehicle.
6. The method of claim 2 , wherein the planning decision plans a future trajectory of the vehicle.
7. The method of claim 1 , wherein the agent comprises a robot.
8. The method of claim 1 , wherein processing the fused representation to generate the estimated position in the three-dimensional region of the environment comprises:
processing the fused representation using a pose estimation neural network that has been trained on training data comprising both (i) labeled point cloud data that associates each of multiple point clouds with corresponding human assigned keypoints and (ii) unlabeled point cloud data for which human assigned keypoints are unavailable.
9. A system comprising one or more computers and one or more storage devices storing instructions that, when executed by the one or more computers, cause the one or more computers to perform operations comprising:
generating a fused representation of an image of an environment and a point cloud of a three-dimensional region of the environment, wherein the point cloud comprises a plurality of data points, and wherein the generating comprises:
generating, based on the image and for each of a plurality of keypoints, a score for each of a plurality of locations in the image that represents a likelihood that the keypoint is located at the location in the image; and
generating, for each of the plurality of data points in the point cloud, a respective feature vector that includes at least some of the scores generated based on the image;
processing the fused representation to generate, for each of the plurality of keypoints, an estimated position in the three-dimensional region of the environment; and
controlling an agent based at least on the estimated position for each of the plurality of keypoints.
10. The system of claim 9 , wherein controlling the agent based at least on the estimated position for each of the plurality of keypoints comprises:
generating, based at least on the estimated position for each of the plurality of keypoints, a planning decision for the agent; and
controlling the agent to implement the planning decision by transmitting one or more electronic signals that have been generated in accordance with the planning decision to one or more control units of the agent.
11. The system of claim 9 , wherein the agent comprises a vehicle, and wherein the environment is an environment in a vicinity of the vehicle.
12. The system of claim 9 , wherein the plurality of keypoints collectively define an estimated pose of each of one or more pedestrians in the environment.
13. The system of claim 11 , wherein the vehicle comprises an autonomous vehicle, or a semi-autonomous vehicle.
14. The system of claim 10 , wherein the planning decision plans a future trajectory of the vehicle.
15. The system of claim 9 , wherein the agent comprises a robot.
16. The system of claim 9 , wherein processing the fused representation to generate the estimated position in the three-dimensional region of the environment comprises:
processing the fused representation using a pose estimation neural network that has been trained on training data comprising both (i) labeled point cloud data that associates each of multiple point clouds with corresponding human assigned keypoints and (ii) unlabeled point cloud data for which human assigned keypoints are unavailable.
17. One or more non-transitory computer-readable storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:
generating a fused representation of an image of an environment and a point cloud of a three-dimensional region of the environment, wherein the point cloud comprises a plurality of data points, and wherein the generating comprises:
generating, based on the image and for each of a plurality of keypoints, a score for each of a plurality of locations in the image that represents a likelihood that the keypoint is located at the location in the image; and
generating, for each of the plurality of data points in the point cloud, a respective feature vector that includes at least some of the scores generated based on the image;
processing the fused representation to generate, for each of the plurality of keypoints, an estimated position in the three-dimensional region of the environment; and
controlling an agent based at least on the estimated position for each of the plurality of keypoints.
18. The non-transitory computer-readable storage media of claim 17 , wherein controlling the agent based at least on the estimated position for each of the plurality of keypoints comprises:
generating, based at least on the estimated position for each of the plurality of keypoints, a planning decision for the agent; and
controlling the agent to implement the planning decision by transmitting one or more electronic signals that have been generated in accordance with the planning decision to one or more control units of the agent.
19. The non-transitory computer-readable storage media of claim 17 , wherein the agent comprises a vehicle, and wherein the environment is an environment in a vicinity of the vehicle.
20. The non-transitory computer-readable storage media of claim 17 , wherein the plurality of keypoints collectively define an estimated pose of each of one or more pedestrians in the environment.