The First “Self-Driving” Telescope Goes Stargazing (SkAI )

0:00 Anna: Happy horse's birthday. Yes. Today the

0:03 1st of August, every horse in the Southern

0:06 hemisphere is officially a year older,

0:08 whether it was born yesterday or a decade

0:11 ago. And in the spirit of a fresh start,

0:14 our lead story today is a telescope that's

0:17 learning to drive itself.

0:19 Avery: A telescope that decides all on its own where

0:22 to point next. In real time, reading the

0:25 weather and the moonlight high in the Chilean

0:27 Andes, we'll get into exactly what that means

0:30 and whether we should be thrilled or a little

0:33 bit nervous about it on,

0:34 Anna: um, the three of the biggest stories from the

0:36 week just gone. Beetlejuice finally

0:39 caught with a companion. A bizarre three

0:42 lobed asteroid with its own little moon

0:45 and a black hole flexing its muscle across

0:48 300,000 light years.

0:51 Avery: And a, uh, look ahead at what is frankly a

0:53 blockbuster August in the sky. This is the

0:56 weekend wrap. Let's go.

0:58 Anna: So picture the job of an observing astronomer

1:01 for a moment. You've got one night on a world

1:04 class telescope. The clock is ticking and

1:06 every hour is precious. But the sky

1:09 doesn't sit still for you. Clouds roll in,

1:12 the moon rises and washes out the faint

1:14 stuff, the air steadies or turns

1:17 turbulent and you're constantly making

1:19 judgment calls. Point here now, save

1:22 that target for later. Skip this one

1:25 entirely. It's a balancing act astronomers

1:28 have honed over decades.

1:30 Avery: And it's exactly the kind of judgment we tend

1:32 to assume needs a human.

1:33 Which is what makes this story land. A team

1:36 from Northwestern University, the University

1:39 of Chicago and Firma Lab working through a

1:41 research center with the rather lovely name

1:44 of the AI Institute for the sky

1:46 sky, pronounced sky, have built a

1:49 deep learning system that makes those calls

1:52 itself in real time.

1:54 Anna: And they didn't test it on a toy. They put it

1:56 in charge of scheduling for the dark energy

1:59 camera. That's a 570

2:02 megapixel monster of an instrument

2:04 mounted on the Victor M. Blanco 4 meter

2:07 telescope at the Cerro Tololo Inter American

2:10 Observatory in Chile. That's one of the

2:13 great observing sites on Earth up in the

2:16 Andes, far from city light.

2:18 Avery: And this is the part I want to sit on because

2:20 it's the leap the, the AI didn't just

2:22 generate a plan for the night in advance and

2:24 hand it over. It adapted that plan on the

2:27 fly as uh, the conditions changed around it,

2:30 the weather shifts, the moonlight changes and

2:32 the system redecides where to look next.

2:37 Anna: They didn't sit down and program in. All the

2:39 rules astronomers have worked out over the

2:41 years. Don't point near the Moon, prioritize

2:44 this when the seeing is Good. And so on. They

2:47 let the model learn on its own. Alex

2:50 Drelica Wagner at UH UChicago, who co led

2:53 the project, described the method really

2:55 simply. They trained it on years of

2:58 historical observations from the Dark Energy

3:00 Survey, showing it where the telescope was

3:03 pointing at one moment and asking it to

3:05 predict

3:06 Avery: the next move and then correcting it.

3:09 So it's the classic learning loop guess

3:12 compared to what the human astronomers

3:14 actually did. Adjust. Do that across

3:17 years of real observing nights and you end up

3:19 with a system that's absorbed the craft

3:21 rather than being told the recipe and the

3:24 headline result.

3:25 Anna: Across two full observing runs, it

3:28 performed comparably to human schedulers.

3:31 It implicitly accounted for the atmospheric

3:33 conditions and the moonlight without anyone

3:36 ever writing those as explicit rules.

3:38 Drelika Wagner called it an important

3:41 milestone towards more autonomous

3:43 observatories and noted the real

3:45 achievement was setting up all the

3:47 infrastructure to actually deploy it on a

3:50 national observatory. That's the difference

3:53 between a nice demo and a working tool.

3:56 Avery: So let's talk about why this matters beyond

3:58 the neat M headline, because I think it's

4:00 genuinely a sign of where astronomy is

4:02 heading. We're entering the era of survey

4:05 astronomy at a scale that's honestly hard to

4:08 picture. The Vera Rubin Observatory, also

4:11 in Chile, is going to image the entire

4:13 southern sky every few nights, over, over and

4:16 over for a decade.

4:18 Anna: And the data volume from that is staggering.

4:20 The kind of fire hose where you simply cannot

4:23 have a human hand tuning every decision fast

4:25 enough to keep up. If you want to catch the

4:28 things that change. A supernova flaring, an

4:30 asteroid drifting through something that

4:32 wasn't there last night. The telescope has to

4:35 be nimble. It has to react.

4:37 Avery: Which is exactly what a self scheduling

4:39 system buys you. Speed and consistency

4:42 and the ability to respond to the sky as it

4:45 actually is on the night, rather than as you

4:47 guessed it would be a week ago when you wrote

4:49 your plan.

4:50 Anna: There's a workload angle too, and I don't

4:52 want to skate past it. Observing runs are

4:55 exhausting. Astronomers up all night making

4:58 hundreds of small decisions under pressure,

5:00 automating the routine scheduling frees

5:03 people up for the science, the

5:04 interpretation, the creative questions that

5:07 machines are nowhere near touching.

5:09 Avery: Now, the honest tension, because whenever we

5:12 say self driving anything, a reasonable

5:14 person's eyebrows go up. Are we handing the

5:17 keys to the machine? And I think the fair

5:20 answer here is not really. Not in the way

5:22 that phrase suggests. This is a system

5:25 deciding the order and targeting of

5:27 observations. Which patch of sky in which

5:29 filter when within goals that

5:32 humans set, the scientists still decide what

5:35 the survey is for.

5:36 Anna: That's the key distinction. The why

5:39 stays human. The where next right

5:42 now is what's being handed over. And it's

5:44 being handed over to something that learned

5:46 from the very astronomers it's now standing

5:49 in for. So it's less a replacement and more

5:52 a, uh, very fast, very tireless apprentice.

5:54 Avery: And for our listeners who observe themselves,

5:57 and I know a lot of you do, there's something

5:59 quietly relatable in this. Every one of you

6:02 has stood in the backyard doing this same

6:04 calculation. Is it worth setting up tonight?

6:06 Is that haze going to clear? Should I chase

6:09 the faint galaxy now or wait for it to climb

6:12 higher? This AI is doing your

6:14 backyard math just, uh, at 4 meters of

6:17 aperture and a few hundred megapixels.

6:19 Anna: And I love that it's happening in the

6:21 southern hemisphere at Cerro Tololo on a

6:24 telescope that's been a workhorse of southern

6:26 sky astronomy for decades. The Blanco helped

6:29 make the original dark energy discovery. Now

6:32 it's helping pioneer how the next generation

6:35 of telescopes will run themselves. That's a

6:37 nice bit of continuity.

6:39 Avery: It is. So the first self driving telescope

6:41 has gone stargazing, and it drove well. Keep

6:44 an eye on this one, because the observatories

6:46 of the2030s are going to be built around

6:49 exactly this idea.

6:51 Now into our three highlights from the week.

6:53 And if you missed Wednesday's episode, this

6:55 was a big one. Betelgeuse, that famous red

6:58 orange shoulder of Orion, turns out not to be

7:01 alone.

7:01 Anna: And for a century, people have suspected a

7:04 companion star tugging at it. Because

7:06 Betelgeuse has this long, slow rhythm in its

7:09 brightness that a single star struggles to

7:11 explain. But nobody had ever seen the

7:14 companion directly. It's lost in the glare of

7:17 a supergiant that could swallow the orbit of

7:19 Jupiter.

7:20 Avery: Until now, a team led by Miguel

7:22 Montajes, using the European Southern

7:25 Observatory's Very Large Telescope in Chile,

7:28 got the clearest image yet of what's very

7:30 likely Betelgeuse B, a faint

7:33 companion riding close. Montarge called it,

7:36 uh, the conclusion of a century long quest,

7:38 which is not a phrase astronomers throw

7:41 around lightly.

7:42 Anna: And it's a lovely reminder that even the

7:44 stars we think we know best still have

7:46 secrets. Betelgeuse is one of the most

7:49 studied stars in the sky, and it was quietly

7:52 hiding a partner in plain sight. If you want

7:55 the full detail, it's all in Wednesday's

7:57 episode 153. Highlight 2. And this

8:00 one's brand new. We ran it just yesterday.

8:03 Say hello to asteroid 44 Nisa,

8:05 which has turned out to be genuinely strange

8:08 in the best way.

8:09 Avery: Strange how? Give people the picture.

8:11 Anna: So most asteroids we imagine as a single

8:14 lump or maybe a two part contact binary,

8:17 like a peanut. NISA is the first

8:20 known tri lobe asteroid. Three

8:22 distinct lobes joined together by narrow

8:25 necks, three connected blobs, and

8:28 sophisticated high contrast imaging from the

8:31 Large Binocular Telescope in Arizona and the

8:34 VLT in Chile is what pulled the shape out of

8:37 the glare.

8:37 Avery: And it's not traveling alone either, is it?

8:39 Anna: Uh, it is not. They also spotted a small

8:42 moon roughly a kilometer across,

8:45 orbiting it. And that little moon is

8:47 scientifically gold because watching how it

8:50 orbits lets you weigh the asteroid. You get

8:53 its mass, and from the shape, you get its

8:55 volume. And together those give you density,

8:58 which tells you what it's actually made of.

9:00 So a weird shape plus a handy moon equals

9:03 a real physics windfall. Full stories in

9:06 yesterday's episode 155 and

9:09 Avery: highlight three takes us right up in scale

9:11 from a kilometer wide moon to something

9:13 spanning 300,000 light years. On

9:16 Thursday, we covered new X ray observations

9:18 of a quasar called H

9:20 1821643

9:22 Anna: and a quasar for anyone just joining us

9:25 is a supermassive black hole in the middle of

9:28 a galaxy, feeding so voraciously that the

9:31 surrounding material blazes brighter than the

9:33 whole galaxy around it.

9:35 Avery: Exactly. And using the Xrism M X

9:38 ray telescope, astronomers traced winds

9:40 driven by that black hole, reshaping the gas

9:43 across an entire galaxy cluster.

9:45 Turbulence stirred up over hundreds of

9:47 thousands of light years. One black hole at

9:50 the center, setting the weather for a whole

9:52 cluster of galaxies.

9:53 Anna: It's one of those numbers that resets your

9:55 sense of scale. The reach of a single

9:58 object from a point out across a distance

10:01 that three times the width of our entire

10:03 galaxy. And it's a nice

10:05 Xrism callback. That telescope

10:08 keeps earning its keep. That was Thursday's

10:11 episode 154. Quick Community

10:14 Note before we look up. If you're enjoying

10:16 the wrap, the new astronomydaily

10:18 IO is the place to live. The full back

10:21 catalog's there. A, uh, news feed that

10:23 updates through the day and night. Listener

10:26 reviews and a, uh, newsletter sign up. So the

10:28 day's stories land in your

10:30 Avery: inbox and it's a contact facility on the site

10:33 now. So if there's a story you think we've

10:35 missed or a bit of sky you want us to talk

10:37 about, tell us directly. We do read them.

10:39 And that brings us to the sky. And August is

10:42 a genuine blockbuster. So this is a preview

10:45 of the whole month, not just tonight.

10:47 Anna: Start with a comet. Comet 10P

10:49 Tempel 2 reaches perihelion its

10:52 closest approach to the sun on 2 August,

10:55 and it's near its peak brightness in early

10:57 August. It's a binocular object rather than a

11:00 naked eye showpiece. So find dark sky and

11:03 sweep for it for us in the Southern

11:05 Hemisphere. It's reasonably placed Northern

11:08 Hemisphere observers. You can catch it too,

11:10 just lower. Check a sky app for your exact

11:13 Avery: local rise time and the headline event of the

11:16 month. And it's a northern hemisphere one. A

11:18 uh, total solar eclipse on the 12th of

11:20 August. The path of totality crosses

11:23 Greenland, Iceland, Spain and Portugal.

11:26 If you're anywhere in Europe, you'll likely

11:28 get at least a partial. And I have to say it

11:30 every single time, never look at the sun

11:32 without proper certified eclipse glasses.

11:35 No exceptions.

11:36 Anna: And here's the beautiful coincidence. That

11:39 Same night, the 12th into the 13th, the

11:41 Perseid meteor shower peaks. And this year it

11:44 peaks under a new moon. No moonlight to wash

11:47 them out at a good dark site. The Perseids

11:50 enthro well over 100 meteors an hour at

11:53 their best with bright fireballs.

11:55 Avery: Now the honest Southern hemisphere notes

11:57 because we always give it to you straight,

11:59 the Perseids are a northern hemisphere

12:02 shower. The radiance up in Perseus

12:04 barely clears the horizon for much of the

12:07 south, so we get far fewer. The

12:09 Northern hemisphere gets the real show this

12:12 time our northern listeners get to dark

12:15 sky after Midnight on the 12th. Look up,

12:18 be patient and let your eyes adapt.

12:20 Anna: And building through the month. A six planet

12:23 lineup strung along the sky. A lovely

12:26 reason to learn the ecliptic. The best

12:29 pairings shift week to week. The pre dawn

12:31 sky is where a lot of the action is with the

12:34 brighter planets. We'll give you the night by

12:36 night detail in the daily episodes as each

12:39 event sharpens up.

12:40 Avery: So a comet at the start, an eclipse and

12:43 meteor double bill in the middle. Planets all

12:46 month. Whichever hemisphere you're in, August

12:48 has something for you.

12:50 Anna: That's the weekend wrap for Saturday 1st

12:53 August. A self driving telescope in the

12:56 Andes. A companion for Beetlejuice. A

12:59 three lobed asteroid with a moon and a

13:01 black hole reaching across a cluster.

13:04 Not a bad week.

13:06 Avery: We're back Monday with a fresh episode. Until

13:08 then, from Anna and me and happy birthday to

13:11 every horse listening.

13:13 Anna: Clear skies,

13:20 Wow.

13:25 Avery: Story soul.