Robotics internships: choose a subsystem and build a useful project

Define what your experiment will count as success

For a navigation project, define the start, goal, permitted movements, obstacles and a timeout. Use the same scenarios for each planner. Record unsuccessful attempts as well as completed paths. Otherwise you can accidentally reward a method for failing on the difficult cases.

The following results are invented to illustrate that problem. Both planners attempt ten scenarios. Planner A finishes eight; Planner B finishes all ten. A path-length average is calculated only where a path was completed.

A shorter average can hide the two hardest failures

Compare the full result before calling A more efficient. The final two scenarios are missing from A’s path-length average, not solved in zero steps.

• B’s overall completed-path average is (8 × 13 + 2 × 23) ÷ 10 = 15. A’s reported 12 excludes both failures.
• On the eight scenarios both finish, A is about 7.7% shorter: (13 − 12) ÷ 13. Comparing 12 with 15 would suggest 20% shorter while comparing different sets of completed tasks.
• Neither average specifies how much a failed task matters. Report completion rate and comparable path lengths separately before choosing a method.

Make the failure reproducible

Build the small experiment, save the scenario definitions and use fixed seeds when randomness is involved. Include a blocked route, a narrow passage and an unreachable goal under your model. State whether a failure means no path found, timeout or a rule violation; those outcomes need different fixes.

Change one assumption, such as observation noise, and show whether the result changes. Do not silently tune the scenarios after seeing which method wins. The report should let another person reproduce both a success and a failure, not just watch the best run.

In an application, identify whether you implemented the planner, configured a library or built the evaluation. Ask which subsystem the intern would own and how it is tested before integration. A simulated result does not establish that a policy is safe on a vehicle or physical robot; keep this portfolio work in simulation.

Before you apply

Eligibility

Waymo describes technical internships for bachelor’s, master’s and PhD candidates. A particular research posting can be narrower.

Timing

Use the term and dates of the actual role; a general robotics job is not necessarily a summer internship.

Location

Waymo describes interns as hosted onsite in a hybrid arrangement. Simulation work does not itself mean remote employment.

Pay

Compare the actual offer’s pay, term and relocation terms. Do not buy a robot to meet an assumed application requirement.

Sources: Waymo early careers

Job titles to search

Perception or localization intern

Estimate what is present or where the system is.

Planning or controls intern

Choose a movement or make the system execute it.

Robotics simulation intern

Build repeatable scenarios and test what succeeds or fails.

Where to look

Search routes, not confirmed vacancies.

Compare the work and try an example

Use a role comparison or worked exercise to choose your preparation before tailoring an application.

• Internship skills: distinguish required, preferred and taught on the job
• Build an economic consulting work sample that another person can reproduce
• Citadel versus Citadel Securities internships: choose the employer and role