Purpose of this page: provide some informal guidance on career development, expectations at a postdoc level in a robotics, machine learning or computer vision research discipline area. This is advice tailored to the current state of this ultra-competitive field, not an aspirational vision for how these fields could actually be run in a better way.
Working assumption: that your immediate career intent is to either a) stay in academia and join faculty eventually or b) work in a technical research-focused role in a tech company / startup (which of course won’t apply to many people). Advice also assumes you’re currently in a research intensive postdoc role, or at least one with substantial research component.
Consequently the main focus of this discussion is on transitioning from a junior researcher who is heavily reliant on their supervisor to a largely independent researcher.
Current context: these fields are currently hyped and generally very well-funded, so if you are completely flexible (location, type of research, academic-industry etc.) getting a job of some description is not challenging, although tenure-track jobs in academia are still quite rare. This situation is unlikely to remain like this in the long term, there are always cycles.
Flexiblity Limitations: if you’re constrained in certain ways (certain geographic constraints, family commitments, don’t want to work in certain sectors e.g. defence or certain countries), then even in the current healthy jobs climate, your options are substantially reduced.
Takeaway for Career Prospects: Consequently, whatever your situation, the more competitive you can be with regards to the following, the better your career prospects.
You’re Not A PhD Student Anymore: a typical PhD student starts off knowing relatively little, and undergoes a rapid learning curve where they pick up lots of skills, knowledge and some wisdom. A top postdoc is expected to continue learning rapidly, and to become substantially more capable than they were as a graduating PhD student.
Compare what you could do starting a PhD to finishing a PhD – and extrapolate that to a few years in a postdoc. If anything, a top postdoc’s learning rate should go up because they are much more effective at managing the process of learning than they were as a fresh PhD student.
Your Supervisor as a Useful Benchmark: academic environments are best if they’re not overly competitive or adversarial. However, it can be very useful from a career development perspective to benchmark yourself against your supervisor, to identify your own areas of weakness and opportunities for targeted development. We’ll come back to this concept in a moment.
Areas of Capability at Postdoc Level: Really there are two primary ones: your
1) technical research capability and
2) your general soft skills.
Technical Skills
Recap of your PhD days: As a PhD student, you likely became a more proficient coder and user of common robotics / machine learning / computer vision libraries. You were likely (but not always) better at this at a technical level than your supervisor by the time you graduated. However, you likely (but not in all cases) got substantial and sustained intellectual guidance from your supervisor.
The Next level as a Postdoc: the key challenge as a postdoc is you should become as proficient, or substantially more proficient, than your supervisor, at a larger range of tasks, rather than just technical ability to implement systems in code. As a research-focused postdoc, this is the primary intellectual focus you have. Your supervisor has a million different things demanding their attention.
For example, driving and leading ideation of research ideas, with a significant “hit rate” – a reasonable percentage of these ideas needs to translate into high impact research breakthroughs and top tier publications. Rapidly writing high quality papers, and eventually, funding proposals. Doing top notch technical research presentations. You should aim to be comparable or better than your supervisor at some of these things.
Where Your Supervisor Will Still Play a Substantial Role: assuming a normal postdoc-supervisor situation where the supervisor is older and more experienced, anything that requires substantial soft skills experience personally and professionally will still involve a substantial training and mentoring influence from the supervisor.
A Guiding Principle: your aim should be to make your supervisor mostly redundant in the ongoing technical research process. Your supervisor should rarely be able to contribute anything substantive, because you’ve anticipated it and are several steps ahead already. Technical research meetings should primarily consist of you checking in with your supervisor, rather than heavy and substantial reworkings of your core research from the ground up led by your supervisor.
Of course, your supervisor will still enjoy the chance to provide some input, but this should be a bonus rather than a core critical component: the success of your research should become increasingly independent of supervisor input.
Soft Skills
Soft skills are things like emotional intelligence, leadership, supervision, collaboration, general management of people. It’s used to effectively manage research teams, team projects, organize events, meetings, start and sustain new initiatives (like say a journal club) and so on.
Having excellent and well practiced soft skills can, up to a point, compensate for shortcomings in technical capability.
For example, if you can effectively supervise and manage research teams of PhDs and undergraduates, then you take load off your supervisor and the group.
But you need to be technically capable enough (authenticity) to do this well – no amount of soft skills will compensate for you not being sufficiently competent and experienced in the technical areas you’re supervising. This is perhaps even truer in industry technical roles – you need to be technically authentic in front of the teams of highly talented people you’re leading.
Soft skills also take a long time to build up. A brilliant technical researcher can do amazing things in a few short years, but it takes years to build up the (sometimes painful) life experience professionally and personally to have a high level of soft skills.
An Important Note on Technical Throughput: A typical good PhD student might publish one high quality lead author paper (primarily their ideas and execution) per year of their PhD. This is nowhere near sufficient for a top tier postdoc, who should be substantially more prolific. This can be achieved through one of two primary mechanisms:
- Individual Technical Track: Becoming very effective and fast at ideating, testing, and executing to completion, high quality innovative research that you lead, and/or
- Soft Skills Version: Becoming very effective at recruiting, leading and managing a research team of PhDs and undergradudates doing research, and/or wider network of collaborative teams locally and globally.
As an example of the individual track, if you have mature experimentation pipelines set up (e.g. a vision front-end and mapping back end, plus standardized evaluation metrics and visualizations, and lots of curated datasets), and have a regular stream of potentially very good ideas, you should be able to rapidly test a bunch of ideas. You need to pivot rapidly and repeatedly on-the-fly as your investigations proceed, with many many iterations, without needing regular intellectual input from your supervisor.
You Need to Be On Target, Often: no-matter how efficient you are, you need your ideas to be good enough and feasible enough that a subset regularly end up going all the way to fruition. This is similar to the Amazon ” Leaders are right a lot” principle – you need to hone that killer instinct for innovative, clever research ideas that are actionable.
You Need to Be Honest With Yourself: performing at a world class level is very challenging, and you’ll only be able to this in narrow areas. You need to constantly self-assess and work out what you are really good at, or are becoming really good at, and what you are plateauing at a “competent” level for. If you’ve given it a good attempt and you’re just not becoming a “natural” at something, then it’s probably not for you.
A Detailed Selfie Snapshot: You can then put together a detailed picture of what you’re world class at, what you’re competent at, and what your weaknesses are, that can:
a) guide what opportunities you pursue and
b) guide how you invest in further development and training.
It’s also important to note that what you’re truly excellent at overlaps, but not entirely, with what you most enjoy, and that these notions can evolve over time.
⚠️ Dangers to Avoid ⚠️: To be a productive postdoc, there are certain things you cannot afford to do regularly, ideally at all:
⚠️ Multi-week/month journeys down a pointless research direction. Not pointless in hindsight (that’s an unfair expectation), but where a reasonably experienced fresh set of eyes could have told you at the start that this was pointless. Your supervisor can do this, but should not have to be doing this regularly – it should be self-managed.
⚠️ Not being ruthlessly focused in your planning of your development pathway, always focusing on the most immediate “quick-fail” outcome that will definitely disprove your idea, or clear the way for further development and investigation. Too often researchers expend substantial effort on something that is 2 or 3 steps down the critical pathway, only to find out later that there’s a showstopper earlier on which renders all their work wasted*.
⚠️ Not being constantly opportunistic and revisiting the big picture – there are multiple ways to achieve breakthroughs, and you need to constantly re-evaluate whether you’re focusing on the right angle or niche, given your progress and interim results. What metrics are you trying to optimize, what aspects of performance are you targeting, and what are you not targeting? A top researcher is always considering all the potential opportunities, and rarely misses one.
⚠️ Following your gut too often. On rare occasions, you’ll have such a strong, “gut feel” or belief in an idea that you want to pursue – and you should! But if you find yourself doing this regularly, with very little or no outcome, that is a clear signal that you need to be more ruthless about evaluating your ability to judge your own ideas, and pickier about what you pursue.
*If you have a very long timing runway in academia (e.g. you’re a tenured professor), then most things are never “wasted” in the long run. But as a postdoc, you can’t afford to take this long term view, because your short term performance is the primary determinant of whether you get to have that longer term career.