Robotic Search for Antarctic Meteorites Patriot Hills 1998 ------------------ Panoramic Camera Performance The Antarctic phase of panoramic camera tests has concluded. This report highlights some of the important experiences and lessons learned in specific areas of the camera design and operation during the field trials in Patriot Hills and Independence Hills. Hardware -------- The panoramic enclosure provided a save and sturdy case for the camera equipment. The space between the top of the lens and bottom of the cylinder mount provided ample room for stacking the filters necessary for sun tracking and other experiments and allowed room to manually mount and unmount those filters. The only part of the camera I was consistently worried about breaking was the glass cylinder. I did not drop the camera the whole time, although I got close to dropping it about three different times. Calibration will be difficult with the camera as it is. The mating between the aluminum shell and glass cylinder allows too much movement of the upper section of the housing and mirror, which can move the mirror location by tens of pixels on the image plane. This renders the calibration moot. Once I noticed that, I stopped saving dewarped images in order to conserve disk space since eventually it will be necessary to dewarp the images independently after localizing the mirror in each image. Software -------- The software used for displaying images and interactively saving images was hacked to automatically save images to files on a keypress with names generated from the system clock. There were still problems with this method. The interactive Windows display and raw image saving functionalities required different pixel representations, and for some reason after several image grabs the framegrabber would return an error on switching modes from 8 to 12 bits and then hang, requiring a reboot. Mean time to failure was about 12 to 15 images, which made acquiring image sequences very difficult. That software was replaced with a console based program which did not display any images at all, and simply streamed images to the hard disk as fast as possible. Across the IDE, it was possible to get about 2 fps. After that, the experiments continued without any hitches. Analysis of data collected in Antarctica will begin shortly and results will be made available as tests progress. Procedure: ---------- The idea of collecting ground truth pose or position without a good dGPS system just didn't fly. I will be able to estimate position for some sequences using triangulation within a square region marked by bamboo canes which I set up in the last week of experimentation. For other sequences, position estimates will be poor and resulting mapping and pose inference will likely be poor as a result. I did have an opportunity to collect short traverses, a grid of panoramas from the Horseshoe Valley, and a suntrace using the dark filters. Images from all three parts of the experiment will be analyzed and reported on later. Optical ------- With small roll and pitch angles (i.e. 10 degrees) the bright Antarctic sun appears in the camera field of view at all times. Reflections of light from the glass cylinder and aluminum housing cause glare in the imaging area, and direct sunlight causes blooming on the CCD. Blooming was also seen in the Atacama camera. The circular polarizer appears to cut glare, but not nearly as well as I had hoped. The linear polarizer cuts some glare, but only in certain areas of the image. In the areas of the image where glare is reduced by the linear polarizer, the LP is more effective than the CP, but then other areas have no glare reduction. An AR coating would obviously help, but having talked to people who can do it, I think that the cost cannot be justified. The bright lighting was difficult to deal with. The exposure was sped up to the fastest shutter and the lens stopped down to the highest f-stop, and there was still too much light to properly expose the CCD. I ended up using an ND 2.0 filter the whole time in order to reduce the intensity of incident light to a manageable level. Thermal ------- The panoramic camera setup had few thermal problems. The heat tape and thermostat system installed by Mike M. worked well to bring the camera to near +10 C and keep it within 2 or 3 degrees of that temperature during all operations. One oversight was that I powered the heater through the PC and therefore I could not warm the camera without turning on the PC. I also had no heaters on the PC, but it would have been a good idea to have PC heaters to bring the PC to operating temperature before booting. I had one day when the hard disk appeared to have problems, but after the computer warmed up and I rebooted it, things appeared OK. I am still not sure that the problem was one of temperature. Thermal issues were most likely assuaged by the unexpectedly mild environmental conditions experienced. The system was a bit underdesigned for the conditions expected, but worked well in the conditions present at Patriot Hills. Because the air temperature was consistently only a few degrees below zero and the sun was so often shining, especially when field work was feasible, the sun would often warm the flight box to well above zero before work began. This may indicate that painting Nomad matte black could greatly reduce concerns of thermal issues even in colder regions. Electrical ---------- The panoramic camera had no electrical problems to speak of. I forgot the power cable when relocating to the Independence Hills moraine and had to rewire the box with a regular power strip, but that was more of a problem of unpacking and repacking than an electrical problem. The 150 W required for operations proved to be a burden. The panoramic camera setup basically requires its own dedicated generator for portability, and we apparently underestimated our needs for multiple generators. A decent lead acid battery would only last for about 1 to 2 hours of field operations before requiring recharging, which was not possible given the needs of the experiment. Nomad would be fine with the power requirements, but a solar powered machine will require a much less resource-hungry design. Data Collection: ---------------- Data collection occurred in several phases. The short expedition to Independence Hills allowed collection of 5 sequences of panoramic images on the blue ice field between the mountains and the moraine. The sequences consist of 16, 32, 6, 16, and 60 images, respectively. The sequences show blue ice and nearby scattered rocks deposited on the ice, a moraine at medium range rich with visual texture, and the far away (1 km) mountains, allowing landmark tracking at many distances. One day was spent dragging the panoramic experiment behind a skidoo and collecting panoramic images on an irregularly shaped grid of positions within the valley between Patriot and Independence Hills. Each image is tagged with GPS from a handheld unit, so accuracy is on the order of 40 meters. These panoramas were taken on the order of 500 meters apart, and the features which persist in the data set will be the Patriot and Independence Hills. From these images, it should be possible to test landmark based position estimation for long range traverses, using far away landmarks. One 24 hour period was spent with the camear in a fixed position with a known orientation. An ND 5.0 filter was placed between the mirror and collecting lens. Once every 15 minutes, the camera would take an image which contains the sun as a white disk and darkness in the remainder of the image. Each image is tagged with the exact time of the image capture. From these images it should be possible to localize the sun, measure the azimuth, and using solar ephemeris calculate the absolute orientation of the camera. The orientation calculated with this method can be compared to a compass bearing taken while the camera was deployed. After code changes were made, eight more sequences were collected, four of which were taken near the CMU camp, and four of which were taken near the Patriot Hills moraine. The CMU camp sequences will allow testing of visual tracking of low contrast features, such as patches of snow. The moraine sequences will allow short, medium, and long range feature tracking as did the other moraine data. Proposed design changes: ------------------------ The panoramic camera that will be deployed in 2000 will need to be smaller, lighter, more rugged, and lower power. I am less concerned with the component design for 1999 since the Nomad platform allows for much more in terms of resources. For many reasons mentioned above, I think that we need to move to a camera with rigid posts for a mirror mount. That design is like the DRES and Columbia cameras. Pros and cons of the post mounting: Pros ---- * Rigid mirror placement. Necessary for calibration since the dewarping LUT is calculated based on the expected mirror location and millimeter differences in actual location cause great changes to the true azimuth and elevation to pixels in the image. * Rugged. The glass cylinder is the most fragile part of the mechanical assembly. * No glass surface to create glare. AR coatings are prohibitively expensive and polarizing filters don't cut glare enough. Cons ---- * Reduced field of view. However, there will be shadowing of the panoramic by other devices on board and the "shadows" are in fixed locations in the field of view and can be dealt with. * Requires mechanical rework. However, in the short term it will only require changes to the very top part of the enclosure since the cylinder and cylinder mount can be removed and replaced leaving the bulk of the existing enclosure in tact. My feelings on the future robot is that the existing hardware is too big, bulky, heavy, and power hungry anyway, and that mechanical rework is inevitible if the camera is to be integrated on a solar machine in under two years. Conclusions ----------- Overall, the performance of the panoramic camera was satisfactory for field tests and data collection. The field experience gained during this field season has provided information which will greatly influence the design of the camera which will be deployed and the vision algorithms which will developed for navigation onboard Nomad in 1999 and the new robot in the third year. The data collected will allow feasibility tests in visual tracking and landmark navigation, sun tracking, and other proposed uses for the panoramic camera. The next phase of testing will commence now using data collected in the field to test the utility of the images for autonomous navigation purposes. Below is a short version of the story: Summary: -------- * Panoramic tests in Antarctica during 1997 field trials offered an opportunity to test hardware, software, optical, thermal, and electrical performance of the panoramic sensor in field deployment under extreme cold and wind conditions, and to collect data for stateside analysis and evaluation of utility of images for Landmark Based Navigation, Solar Visual Compass, and Skyline Based Position Estimation. * Hardware configuration was sufficient from thermal and operational standpoint, but will require changes in order to allow reliable calibration in the field. * Software problems were encountered which apparently have nothing to do with environmental conditions. Fixes were made, and commensurate procedural changes were made to the experiment to facilitate data collection. * Optical arrangement has some problems. Glass cylindrical housing creates high glare. Polarizing filters proved ineffective for removing this problem. Blooming is still a problem with the Antarctic sun always in view, remaining between 13 and 33 degrees above the horizon. * In its current configuration, the panoramic sensor had no thermal problems at all. However, in moving to lower power electronics, the waste heat will no longer be available for thermal control and operating temperature limits may become relevant. * Electrical system worked fine. Lower power version of the sensor will likely be required on the next generation rover, but Nomad's power budget will support the current design. * Overall, the experiment returned over 500 panoramic images. These images include 13 sequences taken near Independence Hills, CMU Base Camp, and Patriot Hills; a series of images sampled on a 500 meter grid in the valley between Independence and Patriot Hills; and one suntrace in which the camera remained fixed for 24 hours with a dark filter imaging the sun once every 15 minutes as it circled overhead. * Environmental conditions have made clear the need for redesign of the sensor for operations in 1999. * Significant field experience was gained through the use of the camera in Antarctic conditions, which has provided a wealth of information on programmatics and human factors issues to be considered in future rover deployments in Antarctica.