Safe traversable area estimation in unstructured free-space using deep convolutional neural network

1. A sidewalk robot comprising:

a LIDAR sensor;

a vehicle system comprising mechanics and a power source to propel the sidewalk robot along a path of an environment;

a processor; and

a non-transitory computer readable medium including instructions that, when executed by the processor, cause the processor to perform operations including:

projecting a laser light from the LIDAR sensor, wherein the projection is directed toward the environment occupied by the sidewalk robot;

receiving one or more reflections of the laser light as LIDAR data for a safe traversable area of the environment that includes a sidewalk, wherein the safe traversable area is absent lane markers and corresponds with a metro geography, and wherein the LIDAR data corresponds with a three-dimensional layout of the environment;

projecting the three-dimensional layout of the environment to one or more two-dimensional frames, wherein the one or more two-dimensional frames correspond with different perspectives associated with the sidewalk robot in the environment;

providing the one or more two-dimensional frames to a trained algorithm, wherein output of the trained algorithm generates a mask identifying the safe traversable area of the environment and a non-traversable area of the environment, the safe traversable area including an area occupied by a dynamic object detected by the sidewalk robot; and

enabling operation of the sidewalk robot using the mask, wherein the operation of the sidewalk robot avoids the non-traversable area of the environment.

2. The sidewalk robot of claim 1 , wherein the operations further include:

training an algorithm to generate the trained algorithm, wherein the training comprises:

determining a plurality of inputs associated with the safe traversable area; and

applying the plurality of inputs to the algorithm to generate the trained algorithm.

3. The sidewalk robot of claim 1 , wherein the trained algorithm is provided to the sidewalk robot when a difference between a first safe traversable area and a second safe traversable area are within a threshold accuracy for the environment.

4. The sidewalk robot of claim 1 , wherein the area occupied by the dynamic object is identified as part of the safe traversable area when a distance between the dynamic object and the sidewalk robot exceeds a threshold distance.

5. The sidewalk robot of claim 1 , wherein the mask identifying the safe traversable area of the environment is a first mask, and the operations further include:

receiving RGB data, wherein the RGB data is received as a two-dimensional frame;

providing the RGB data as input to a second trained algorithm, wherein a second output of the second trained algorithm produces a second mask; and

combining the first mask and the second mask to generate a combined mask identifying the safe traversable area of the environment and the non-traversable area of the environment.

6. The sidewalk robot of claim 1 , wherein the operation of the sidewalk robot using the mask avoids using a high-definition (HD) map or a global positioning system (GPS).

7. A non-transitory computer-readable storage medium storing instructions executable by one or more processors, wherein the instructions, when executed by the one or more processors, cause the one or more processors to perform operations including:

projecting a laser light from a LIDAR sensor of a sidewalk robot, wherein the projection is directed toward an environment occupied by the sidewalk robot;

receiving one or more reflections of the laser light as LIDAR data for a safe traversable area of the environment that includes a sidewalk, wherein the safe traversable area is absent lane markers and corresponds with a metro geography, and wherein the LIDAR data corresponds with a three-dimensional layout of the environment;

projecting the three-dimensional layout of the environment to one or more two-dimensional frames, wherein the one or more two-dimensional frames correspond with different perspectives associated with the sidewalk robot in the environment;

providing the one or more two-dimensional frames to a trained algorithm, wherein output of the trained algorithm generates a mask identifying the safe traversable area of the environment and a non-traversable area of the environment, the safe traversable area including an area occupied by a dynamic object detected by the sidewalk robot; and

enabling operation of the sidewalk robot using the mask, wherein the operation of the sidewalk robot avoids the non-traversable area of the environment.

8. The non-transitory computer-readable storage medium of claim 7 , wherein the operations further include:

training an algorithm to generate the trained algorithm, wherein the training comprises:

determining a plurality of inputs associated with the safe traversable area; and

applying the plurality of inputs to the algorithm to generate the trained algorithm.

9. The non-transitory computer-readable storage medium of claim 7 , wherein the trained algorithm is provided to the sidewalk robot when a difference between a first safe traversable area and a second safe traversable area are within a threshold accuracy for the environment.

10. The non-transitory computer-readable storage medium of claim 7 , wherein the area occupied by the dynamic object is identified as part of the safe traversable area when a distance between the dynamic object and the sidewalk robot exceeds a threshold distance.

11. The non-transitory computer-readable storage medium of claim 7 , wherein the mask identifying the safe traversable area of the environment is a first mask, and the operations further include:

receiving RGB data, wherein the RGB data is received as a two-dimensional frame;

providing the RGB data as input to a second trained algorithm, wherein a second output of the second trained algorithm produces a second mask; and

combining the first mask and the second mask to generate a combined mask identifying the safe traversable area of the environment and the non-traversable area of the environment.

12. The non-transitory computer-readable storage medium of claim 7 , wherein the operation of the sidewalk robot using the mask avoids using a high-definition (HD) map or a global positioning system (GPS).

13. A computer-implemented method comprising:

projecting a laser light from a LIDAR sensor of a sidewalk robot, wherein the projection is directed toward an environment occupied by the sidewalk robot;

receiving one or more reflections of the laser light as LIDAR data for a safe traversable area of the environment that includes a sidewalk, wherein the safe traversable area is absent lane markers and corresponds with a metro geography, and wherein the LIDAR data corresponds with a three-dimensional layout of the environment;

projecting the three-dimensional layout of the environment to one or more two-dimensional frames, wherein the one or more two-dimensional frames correspond with different perspectives associated with the sidewalk robot in the environment;

providing the one or more two-dimensional frames to a trained algorithm, wherein output of the trained algorithm generates a LIDAR mask generated from LIDAR data identifying the safe traversable area of the environment and a non-traversable area of the environment;

generating a combined mask by fusing the LIDAR mask with a RGB mask generated from RGB data; and

enabling operation of the sidewalk robot using the combined mask, wherein the operation of the sidewalk robot avoids the non-traversable area of the environment.

14. The computer-implemented method of claim 13 , further comprising:

training an algorithm to generate the trained algorithm, wherein the training comprises:

determining a plurality of inputs associated with the safe traversable area; and

applying the plurality of inputs to the algorithm to generate the trained algorithm.

15. The computer-implemented method of claim 13 , wherein the trained algorithm is provided to the sidewalk robot when a difference between a first safe traversable area and a second safe traversable area are within a threshold accuracy for the environment.

16. The computer-implemented method of claim 13 , wherein the LIDAR mask identifies both dynamic objects and static objects.

17. The computer-implemented method of claim 13 , further comprises:

receiving the RGB data as a two-dimensional frame; and

providing the RGB data as input to a second trained algorithm, wherein a second output of the second trained algorithm produces the RGB mask.

18. The computer-implemented method of claim 13 , wherein the operation of the sidewalk robot using the fused mask avoids using a high-definition (HD) map or a global positioning system (GPS).