Inside Unmanned Systems provides actionable business intelligence to decision-makers and influencers operating within the global UAS community. Features include analysis of key technologies, policy/regulatory developments and new product design.
Issue link: https://insideunmanned.epubxp.com/i/813246
40 unmanned systems inside April/May 2017 LAND AGRICULTURE don't want to even give it a path—we could just tell it to go to plot number 15 and it'd fig- ure it out on its own." Chowdhary noted that the active radiation sensors "don't really work well with a lot of sunlight." One way around that is to work at night, he said. The robot transmits its data to the re- searcher's computer in real time. The scien- tists are developing software to construct 3-D images of each plant based on this data. They are also working on computer models to identify what genes are linked with specific physical traits, predict how the plants might grow and develop over time and assess which plants should be bred together. The scientists are currently using the robot to analyze fields of sorghum. The goal is to se- lect the sorghum lines that are most produc- tive and efficient at generating biomass that can be converted to biofuels such as ethanol. " OVER TIME, WE DON'T WANT TO EVEN GIVE IT A PATH—WE COULD JUST TELL IT TO GO TO PLOT NUMBER 15 AND IT'D FIGURE IT OUT ON ITS OWN." Girish Chowdhary, director, Distributed Autonomous Systems Lab On-the-Ground Challenges By training, Chowdhary is an aerospace engi- neer who spent 10 years developing more than 15 research unmanned aerial vehicles (UAVs). He noted that working on a rover posed intrigu- ing challenges compared to developing drones. " With UAVs, perception a nd nav iga- tion problems are simplif ied in the air," Chowdhary said. "But when you're close to the ground, operating close to crops, it's dif- ficult to make rugged equipment that works for a long time outdoors in a harsh, uncertain environment on its own without supervision. There are all kinds of things that can happen on a farm—say, a puddle is in your way—that you have to think about when designing their decision-making systems." One strategy the researchers are pursuing to deal with these challenges involves algo- rithms that can learn quickly from field data to adapt to uncertain environments. "We're essentially doing AI (artificial intelligence) in the real world," Chowdhary said. For instance, the scientists are developing learning-based adaptive models to improve robot performance across the range of fric- tion the robot's treads might encounter, such as that from mud, snow, grass or gravel, Chowdhary said. The researchers are also de- veloping models that can predict how a field might evolve over time to, for instance, better predict the best path through that field when it has just rained. The AI algorithms are utilizing data from the sensors to track user defined paths. Their main objective is to provide very high preci- sion position tracking by mitigating track-slip in muddy fields. In addition to surface condi- tions, online parameter estimation is required to minimize deviations between the system model and real-time system. "In practice, it is difficult to model a system perfectly, because there are always uncertain- ties, disturbances and unmodeled dynamics. Adaptive models are necessary for online The rover should per form equally well ana ly zing the phenot y pes of other ta ll- growing row crops such as corn and wheat, Chowdhary said. "If you look at the trajec- tory of agriculture in the United States, it's succeeded in growing massive amounts of food with very few people. The average size of the American farm is 800 acres, and you now typically only have four people manag- ing that farm," Chowdhary said. "That's due to a lot of efficient mechanization. My hope is that we can continue to make agriculture more and more efficient, to grow more and more food with less and less effort from the human side."