Presentation is loading. Please wait.

Presentation is loading. Please wait.

Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 Science Planner Science Observer Life in the Atacama Design Review December 19, 2003.

Similar presentations


Presentation on theme: "Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 Science Planner Science Observer Life in the Atacama Design Review December 19, 2003."— Presentation transcript:

1 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 Science Planner Science Observer Life in the Atacama Design Review December 19, 2003 Trey Smith (many ideas from Dave W, Mike, Dom, others)

2 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 2 Science Observer

3 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 3 (Notes for previous slide) The side view (at top of slide) shows repeated nav cam views as the rover moves forward. The top view (at bottom of slide) shows the same field of view from above, then shows the map generated by the science observer. The map includes rocks above a certain size threshold (segmented from the images), and a characterization of the soil types in different grid cells (e.g., color and texture of the soil).

4 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 4 Experiment Generator Spectrum, $30 FI image set, $100

5 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 5 (Notes for previous slide) The experiment generator starts with the map generated by the science observer. Based on this map and on scientist preferences, it generates a prioritized list of experiments. An experiment is a feature and a specific data set to gather about that feature. The priority of each experiment is expressed a reward (the number of “virtual dollars” the rover would get for achieving that goal).

6 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 6 Science Planner Spectrum, $30 Warm up Calibrate Take reading 3 minutes, 2000 J FI image set, $100 Calibrate Spray dyes 7 channels 30 minutes, 45,000 J

7 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 7 (Notes for previous slide) The science planner receives experiments from the experiment generator. For each experiment, it generates a plan for taking the relevant data. Associated with each plan is the amount of time and energy it will consume. The planner also receives constraints from the mission plan. These include hard constraints (e.g., can't take longer than 5 minutes) and soft constraints (e.g., only execute experiments if they are worth more than $10/minute). Based on these constraints, the planner chooses a subset of experiments to execute.

8 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 8 Key Requirements Baseline (no science autonomy) Science observer None – present for science autonomy only Science planner Generate sequences for regular sampling strategies during site survey (and traverse?) Perform supporting actions, e.g. warming up and calibrating instruments Act within constraints of mission plan With science autonomy “Requirements by design”

9 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 9 Architecture Science Observer Science Map Preference Database Rover Executive rocks, soil units SPI, nav, spectra, etc. Instrument Controllers Instrument Controllers Instrument Controllers Science Planner initiate planning / submit plan notify about special features

10 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 10 Baseline Design Generate sequences for regular sampling Experiment generator outputs uniform-reward goals on grid Perform supporting actions, e.g. warm-up Probably use EUROPA Act within constraints of mission plan Tag actions with position, time, and energy Simulate plan forward to determine validity If no valid plans, give up and advance to next mission- level objective

11 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 11 Implementation Issues Early steps Determine interface with IDEA Preliminary models of science actions Develop test problems based on ops concept Risks Planner-based systems often difficult to understand, mitigate by As much testing as possible Training users New modules more likely to have unexpected problems, mitigate by Incremental descoping

12 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 12 Science Map Rocks Size Albedo

13 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 13 (Notes for previous slide) The image at the left is a top-down view of a 4 m x 4 m area we are analyzing for science. The science observer segments out rocks from the background. Associated with each rock are some characteristics, including size, albedo, color, vis/NIR spectrum, texture, angularity, etc. At the right, we have positioned the different rocks along just two of these axes, albedo and size. In order to organize this data, we run a clustering algorithm that associates similar rocks. The clusters are useful for identifying novelty (rock not in an existing cluster) and for scientist preferences (“I want more data about type 3 rocks”).

14 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 14 Science Map Soil Units Texture Color

15 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 15 (Notes for previous slide) Again, the map at the left is a top-down view of a 4 m x 4 m area we are analyzing for science. We have laid down a 1 m resolution grid. The science observer collects data about “soil units”, basically grid cells at that size scale. Soil units are characterized by albedo, color, texture, vis/NIR spectrum, etc. As with rocks, at the right we position the soil units along just two of these “axes”, color and texture. Again, we cluster similar soil units together.

16 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 16 Science Map Rock Distribution Qty 0 6 3 Region Characterization Soil Unit Distribution Qty 0 22 0

17 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 17 (Notes for previous slide) Again, the map at the left is a top-down view of a 4 m x 4 m area we are analyzing for science. The green rectangle at the lower-left of the map is a 2 m x 2 m sub-region we want to characterize. Characteristics of this sub-region are its distribution of rock types, and its distribution of soil units. These are shown in the histograms at the right.

18 Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 18 What Makes Features Interesting Rocks and soil units Specific scientist preferences Based on clusters White rocks, chlorophyll signature, etc. Novelty – not in any existing cluster Region (distribution of rocks and soil units) Homogeneity Hard to learn from a jumble of rocks Geological boundary Adjacent regions with different distributions


Download ppt "Carnegie Mellon Life in the Atacama, Design Review, December 19, 2003 Science Planner Science Observer Life in the Atacama Design Review December 19, 2003."

Similar presentations


Ads by Google