Will More Powerful Processors Super-Charge NASA's Mars Rovers? (utexas.edu) 27
The Texas Advanced Computer Center talks to Masahiro (Hiro) Ono, who leads the Robotic Surface Mobility Group at NASA's Jet Propulsion Laboratory which led all the Mars rover missions (also one of the researchers who developed the software that allows the current rover to operate):
The Perseverance rover, which launched this summer, computes using RAD 750s — radiation-hardened single board computers manufactured by BAE Systems Electronics. Future missions, however, would potentially use new high-performance, multi-core radiation hardened processors designed through the High Performance Spaceflight Computing project. (Qualcomm's Snapdragon processor is also being tested for missions.) These chips will provide about one hundred times the computational capacity of current flight processors using the same amount of power. "All of the autonomy that you see on our latest Mars rover is largely human-in-the-loop" — meaning it requires human interaction to operate, according to Chris Mattmann, the deputy chief technology and innovation officer at JPL. "Part of the reason for that is the limits of the processors that are running on them. One of the core missions for these new chips is to do deep learning and machine learning, like we do terrestrially, on board. What are the killer apps given that new computing environment...?"
Training machine learning models on the Maverick2 supercomputer at the Texas Advanced Computing Center (TACC), as well as on Amazon Web Services and JPL clusters, Ono, Mattmann and their team have been developing two novel capabilities for future Mars rovers, which they call Drive-By Science and Energy-Optimal Autonomous Navigation.... "We'd like future rovers to have a human-like ability to see and understand terrain," Ono said. "For rovers, energy is very important. There's no paved highway on Mars. The drivability varies substantially based on the terrain — for instance beach versus bedrock. That is not currently considered. Coming up with a path with all of these constraints is complicated, but that's the level of computation that we can handle with the HPSC or Snapdragon chips. But to do so we're going to need to change the paradigm a little bit."
Ono explains that new paradigm as commanding by policy, a middle ground between the human-dictated: "Go from A to B and do C," and the purely autonomous: "Go do science."
Commanding by policy involves pre-planning for a range of scenarios, and then allowing the rover to determine what conditions it is encountering and what it should do. "We use a supercomputer on the ground, where we have infinite computational resources like those at TACC, to develop a plan where a policy is: if X, then do this; if y, then do that," Ono explained. "We'll basically make a huge to-do list and send gigabytes of data to the rover, compressing it in huge tables. Then we'll use the increased power of the rover to de-compress the policy and execute it." The pre-planned list is generated using machine learning-derived optimizations. The on-board chip can then use those plans to perform inference: taking the inputs from its environment and plugging them into the pre-trained model. The inference tasks are computationally much easier and can be computed on a chip like those that may accompany future rovers to Mars.
"The rover has the flexibility of changing the plan on board instead of just sticking to a sequence of pre-planned options," Ono said. "This is important in case something bad happens or it finds something interesting...." The efforts to develop a new AI-based paradigm for future autonomous missions can be applied not just to rovers but to any autonomous space mission, from orbiters to fly-bys to interstellar probes, Ono says.
Training machine learning models on the Maverick2 supercomputer at the Texas Advanced Computing Center (TACC), as well as on Amazon Web Services and JPL clusters, Ono, Mattmann and their team have been developing two novel capabilities for future Mars rovers, which they call Drive-By Science and Energy-Optimal Autonomous Navigation.... "We'd like future rovers to have a human-like ability to see and understand terrain," Ono said. "For rovers, energy is very important. There's no paved highway on Mars. The drivability varies substantially based on the terrain — for instance beach versus bedrock. That is not currently considered. Coming up with a path with all of these constraints is complicated, but that's the level of computation that we can handle with the HPSC or Snapdragon chips. But to do so we're going to need to change the paradigm a little bit."
Ono explains that new paradigm as commanding by policy, a middle ground between the human-dictated: "Go from A to B and do C," and the purely autonomous: "Go do science."
Commanding by policy involves pre-planning for a range of scenarios, and then allowing the rover to determine what conditions it is encountering and what it should do. "We use a supercomputer on the ground, where we have infinite computational resources like those at TACC, to develop a plan where a policy is: if X, then do this; if y, then do that," Ono explained. "We'll basically make a huge to-do list and send gigabytes of data to the rover, compressing it in huge tables. Then we'll use the increased power of the rover to de-compress the policy and execute it." The pre-planned list is generated using machine learning-derived optimizations. The on-board chip can then use those plans to perform inference: taking the inputs from its environment and plugging them into the pre-trained model. The inference tasks are computationally much easier and can be computed on a chip like those that may accompany future rovers to Mars.
"The rover has the flexibility of changing the plan on board instead of just sticking to a sequence of pre-planned options," Ono said. "This is important in case something bad happens or it finds something interesting...." The efforts to develop a new AI-based paradigm for future autonomous missions can be applied not just to rovers but to any autonomous space mission, from orbiters to fly-bys to interstellar probes, Ono says.
Re: (Score:3, Funny)
First jobs to be outsourced to Mars.
Re: Could allow for reduced staff, $$$ savings (Score:2)
Or those people can work on another project.
Don't overdo it (Score:2)
"There is no master but Master, and QT-1 is His prophet."
Super-charged rovers (Score:2)
They can do a 10 minute quarter mile
Re: (Score:3)
Re:Grumble grumble grumble (Score:5, Insightful)
The problem is not the cost of human attention or decisions. The problem is the latency. The best case is a bit more than three minutes; the worst case is almost 7.5 times the distance, except then Mars is on the far side of the Sun from Earth, so you really have to wait months for the planets to move so we have a better line of sight. A lot can happen to the rover in that time.
Nice idea. What is the lead time? (Score:5, Informative)
How long will it take to verify that these candidate processors really are rad-hardened enough to live through the trip and operate on Mars? The RAD750 is a 19-year-old chip based on a 23-year-old processor (the PowerPC 750), and part of what makes it robust to single-event upsets (SEUs) is its lithography. More modern silicon uses much smaller circuit elements, but this makes them more susceptible to corruption from high-energy particles in addition to making them faster and use less power.
Perhaps the more cost-effective approach, considering how long it might take to test, build and deploy these chips, is to build in LTE or 5G modems so that the rovers can call up Elon Musk's crew for advice. Those colonists might be on Mars by the time these chips are.
Re: Nice idea. What is the lead time? (Score:2)
Re: (Score:2)
They could also use COTS Intel and AMD CPUs, which are rad-hard processors even if they're not advertised as such. Seven-year-old AMD CPUs (I have the reports in the office, I think it was a 2013 AMD something) will take 1-10Mrad(Si) with no faults, and they've only gotten better since then. Intel are about the same. That's the same as a RAD750 without the cost.
Couldn't find the original paper but here's a presentation on what appears to be the same results [nasa.gov]:
NO processor failures observed (1,4 and 17 Mrad(Si), respectively). âoe17â is NOT a typo.
although Slashdot will no doubt helpfully turn
Re: Nice idea. What is the lead time? (Score:2)
Do the CPUs need special features against single event upset, or can that all be handled externally?
Re: (Score:2)
Radiation problems? (Score:1)
I indeed wonder how those new chips handle the radiation with their smaller etching.
We don't know. Will they ? (Score:2)
If More Powerful Processors will Super-Charge NASA's Mars Rovers, then the headline should be "More Powerful Processors will Super-Charge NASA's Mars Rovers".
Unless Slashdot is run by Clickbait Wankers.
NASA = old (Score:1)
NASA uses old technology .. it's like they have been afraid to innovate since the 1970s .. traumatized by Nixon's policy and budget cut blunders .. and never course corrected from them. Nixon was just as vindictive, even to predecessors, as Trump. He hated that JFK and LBJ got credit for going to the Moon. According to https://www.planetary.org/arti... [planetary.org] Nixon's space policy was:
To treat the space program as one area of domestic policy competing with other concerns, not as a privi
Re: NASA = old (Score:4, Informative)
Re: NASA = old (Score:2)
Nobody wants to lose a $300M spacecraft because a $0.14 transistor from Digikey blew up. Reliability in the mission environment is the #1 consideration. And very few companies have the know-how to harden their products. Itâ(TM)s not easy.
High cost of development + technical challenge + design trade-offs + low volumes = not much competition + lagging performance + high price
Bad idea (Score:2)
I remember the time in the 70's when there were rovers that thought for themselves. We narrowly escaped doom. https://www.youtube.com/watch?v=T6l0Hr3W7aI/ [youtube.com]
Good processors are hard to come by in space (Score:2)
Many space missions use processors that are decades old because they only radiation harden specific processors. Using a modern low power processor would be a dream
Make sure to include a "master" program (Score:1)
As good as the AI allowing the rovers to self navigate and complete more tasks without interaction, it also enables more ways to fail. Just simple things that took humans out of control produced disasters. Look at the recent Boing 737 fiasco where "intelligent" systems actually killed people.
So make sure the AI runs in a sandbox, and do a sanity check for all its requests by an hypervisor / master software. As long as it does not directly control the rover, but gives (hopefully) very useful suggestions we s
They should use the Apple A12X (Score:2)
In case there's any gardens on Mars with walls around them.
Re: (Score:2)
There was a sci-fi short story back in the 60s or 70s, probably in Analog. It was about a semi-autonomous rover on Mars. Turned out there once had been Martians; they were long dead, but their fortifications remained, armed with self-targeting lasers that still worked. One tried to take out the rover, but it got smart and used a hillside for cover, and snuck up behind the laser. If I'm remembering correctly...that was a long time ago. But pretty prescient, I think: Martian rover, with an AI driving it
Rad-Hard-Enough is more than adequate. (Score:5, Interesting)
I worked on a nano-sat project in the late 1990s that, not surprisingly, had a nano budget. That budget most certainly did not include any rad-hard processors. Still, we did all the due diligence needed to learn what was available on the rad-hard market, if only to more firmly exclude it from our plans.
But the project lead, a self-employed consultant and inventor, took that research and asked a simple question: What other products were made on the same production lines as the rad-hard processors, ideally on the same day? We got very lucky: There was just a single vendor of rad-hard parts that also made commercial microprocessors on the same production line on the same days. Just the one on the whole planet.
So we bought a bunch of those micros in the highest bin available, which was for automotive use. Then started the testing needed to see just how rad-hard they were, if at all. First was a gamma test at a local linear accelerator, which the processors passed without any problems.
The next test was to simulate cosmic rays of varying energy, which meant a trip to Brookhaven National Labs on Long Island to use their Tandem 15 MV Van de Graaff heavy-ion accelerators. As luck would have it, our test got to piggy-back on the unused time of another facility user, meaning I'd be running our tests between midnight and 6 AM.
At these energies, all silicon chips will experience severe effects: The key to radiation hardness concerns how the chip recovers from those events. With a minimum of support circuitry and software, the chips survived thousands of events, far more than any mission would be expected to encounter.
The chips were cheap, so we use three of them, just to reduce risks even further. And we did buy a single rad-hard chip, a tiny one that contained only the voting logic for the three chips. It was very inexpensive, about 0.5% the cost of a rad-hard processor.
Just after we had breadboarded all the circuits and software development was well underway we got some bad news that pretty much killed the project: Our ride to space was as a free payload on a Russian launch to Mir, where our nanosats would be released on the way back, soon after leaving Mir. I remember when we got the news that our late-2000 launch date was never going to happen, because Russia would be diverting resources from Mir to ISS.
We tried to get a "space available" ride on a US launcher, but at that time our lack of formal rad-hard paperwork doomed the project. We then tried to share our engineering with other nanosat projects, but found no takers.
The bottom line is that carefully chosen commercial silicon can be just as rad-hard as certified silicon, the main difference being the man-years of paperwork.
However, this was also back in the day when many chip designers still had their own production lines, and there were fewer process changes between different wafer runs: They'd find a process that worked well, then used it for everything. Not the case today, when the vast majority of chip production is outsourced.
I very much doubt it would be possible to repeat today what we did so well back then.
I’m under your control. (Score:1)