On Research Habits and “Taste”

After co-authoring / leading several IROS papers, and then pulling a 40-hour sprint at the end (not recommended), I can finally catch my breath a little.

The main reason I am writing this is that lately I have been thinking a great deal about my students’ research topics, together with some of our recent work. It made me reflect a bit: what does research training actually give you? What exactly is this thing people call “taste”?

First, when I look at many PhD applications, people often go out of their way to emphasize how many papers they have published. And when they graduate, they count again on their fingers: how many papers.

From a practical point of view, is that important? Of course it is. But is it the most important thing? Of course not.

From one angle, I am even a little wary of students with very “rich” research experience. Research has its own routines too, and if one picks up bad habits—or bad ways of thinking about problems—too early, then, unfortunately, it becomes almost impossible to do truly excellent work.

So what do good habits look like?

What follows is only my own view—mine, personally, and only mine.

1. Curiosity

A good researcher is, more often than not, someone who is curious about the world itself.

Curiosity is what gives rise to the desire to explore.

You should first be interested in a broad topic, and then, in the course of the work, keep that interest alive in every detail and every small question.

2. Details

Think deeply about every problem, and keep pursuing it.

To be honest, I am not very good at this myself.

Sometimes I can clearly sense that there is a problem, but then I think, well, the student has not had it easy either… they have already worked really hard… and somehow I let it slide.

The truth is, all of us have inertia. Thinking is tiring.

Especially when a project feels seven- or eight-tenths done and ready to be wrapped up and submitted, we instinctively avoid digging any deeper.

But very often, that is exactly where you are only just beginning to touch what is most central, and most valuable, in the problem.

So I think back to the way my advisor, a few years ago, would just keep asking and asking and asking—he would not stop until I had no answer left.

At the time, I sometimes felt he was simply picking on me for no reason. Incredibly annoying 😅 Only after graduating did I understand.

And now I, too, am on my way to becoming that evil person.

So if nobody is asking you, ask yourself more often:

Does it really work?
What exactly is the novelty, and what exactly is the contribution?

3. Quality

There is an old logic to this world: if you aim high, you may land somewhere in the middle; if you only aim for the middle, you may end up even lower.

One thing I think matters a great deal is this: you have to go all in on the thing you are doing. You have to throw your whole self into it.

Given your current abilities and resources—and yes, deadlines too—you should produce the very best result you can.

At bottom, this is a kind of resolve. You decide that no matter what happens, no matter what goes wrong, you are going to do this work well.

Only with that mindset is it possible, as your abilities grow, for your papers to become better and better, one by one.

4. Persistence

Even in the age of AI, research is still slow work.

I really only have two core requirements for students.

First: show up.

If you are responsible for a project—say we have a weekly or biweekly meeting—I am completely okay with there being no progress. What is not okay is simply not showing up and disappearing.

Unless, perhaps, aliens abducted you because they needed to take you to Mars for a heart-to-heart. Short of that, you show up.

This is almost identical to my view of life: even if the world is a ramshackle stage, the important thing is to get on stage. What does it mean to walk off halfway through?

Second: do not give up lightly.

Once we have chosen something, we keep doing it. If it gets hard, then tough it out, grit your teeth, and keep going. If even that stops working, then change your approach and try again.

Otherwise what? No regrets?

(Of course, all of the above assumes something close to ideal conditions. Reality is full of constraints, compromises, and things one cannot help, and I am not getting into those here.)

Finally, in my few short years of research experience, I have already heard many times that control is dead, SLAM is dead, and, more recently, 3DCV is dead again.

So many things have already died several rounds over… and then come back to life, only to die again.

Perhaps life itself is just a kind of Bayes function.

You have some prior information. You have a current state estimate. You have observations about the world as it is. And you have a great deal of noise that simply cannot be modeled.

From all of that, you make your prediction about the future.

Everyone has different priors. Everyone also has only limited observations, constrained by cognition and by environment. So naturally, people arrive at different forecasts, and different views of what lies ahead.

There is a line that has been popular lately that I like very much:

Your headlights only illuminate fifty meters ahead, but you can still drive the whole way.

Some very senior professors have told me that your research interests will always change, but the person you are is much harder to change.

So, to those of you who are still young: do not be so anxious. Do each thing in front of you, each thing in your hands, as well as you can.

What people call research taste may not be quite as important as it seems.