Job description

Skydio is the leading US drone company and the world leader in autonomous flight, the key technology for the future of drones and aerial mobility. The Skydio team combines deep expertise in artificial intelligence, best-in-class hardware and software product development, operational excellence, and customer obsession to empower a broader, more diverse audience of drone users, from utility inspectors skydio.com to first responders skydio.com, soldiers in battlefield scenarios skydio.com, and beyond skydio.com. About the role: Skydio builds the world’s most advanced autonomous drones used across inspection, public safety, defense, cinematography, and more. Your research won’t languish in a paper: it will fly, shaping how pilots and operators complete real missions in complex environments. Develop and deploy reinforcement learning (and adjacent policy-learning methods) that make Skydio aircraft plan, navigate, and control themselves more intelligently, safely, reliably, and efficiently across our ecosystem: handheld apps, ground control, cloud autonomy services, and fleet workflows. How you'll make an impact: - Navigation & avoidance in the wild: Train policies that adapt online to cluttered 3D scenes (forests, bridges, urban canyons), complementing our stack for robust obstacle avoidance and dynamic goal-seeking. - RL-augmented planning: Fuse learned cost shaping / value functions with trajectory optimization for smooth, agile flight with tight safety envelopes and mission constraints. - Sim to Real at scale: Build scalable datasets and training loops with Isaac Lab, domain randomization, residual learning, and safety filters; validate on real drones weekly. - Human-in-the-loop shared control: Learn assistive policies that blend pilot intent, autonomy priors, and uncertainty-aware behaviors for intuitive control handoffs. - Fleet & multi-agent: Explore decentralized coordination for coverage, pursuit, and collaborative mapping with minimal comms. What makes this internship different: - Real hardware cadence: Prototype in sim, then flight test on production-ready aircraft and edge compute. Your work is meant to make a difference in the real world, not just in simulation. - Safety-first learning: We pair RL with formal constraints, estimation of uncertainty, and conservative action filters for dependable deployment in human environments. - Cross-disciplinary mentorship: Collaborate with perception, mapping, controls, and product teams to close the loop from research to field ops. What makes you a strong fit: - PhD student in Robotics, Machine Learning, Controls, or related field. - Strong fundamentals in RL, control theory, and motion planning; comfort with safety/robustness concepts. - Proficient in Python (PyTorch/JAX/Ray RLlib) and at least one of C++ or CUDA. - Hands-on experience with robotics simulation (Isaac Lab/MuJoCo/PyBullet) and sim2real techniques. - Experience training/deploying policies for navigation, manipulation, or locomotion on real robots or autonomous vehicles. Nice-to-Haves: - Publications (CoRL, ICRA, IROS, RSS, NeurIPS). - Experience with onboard inference optimization (TensorRT, quantization, sparsity). - Familiarity with modern policy learning beyond vanilla RL: diffusion policies, IL/BC, offline RL, model-based RL. - Experience with multi-agent RL or distributed training. Compensation: The hourly rate for this position is $47/hr for Undergrad, $53/hr for Masters, and $70/hr for PhD students*. Compensation will vary based on factors, including skill level, proficiencies, transferable knowledge, and experience. In addition to base salary, Skydio full-time employees are eligible to enroll in our benefit plans and take advantage of a variety of incentives and stipends. *Compensation for certain positions may vary based on the position's location. #LI-DK1 At Skydio we believe that diversity drives innovation. We have created a multidisciplinary environment that embraces the power of diverse perspectives to create elegant solutions for complex problems. We are committed to growing our network of people, programs, and resources to nurture an inclusive culture. Qualified applicants will receive consideration for employment without regard to race, color, religion, sex, sexual orientation, gender identity, national origin, disability, protected veteran status, or other characteristics protected by federal, state or local anti-discrimination laws. For positions located in the United States of America, Skydio, Inc. uses E-Verify to confirm employment eligibility. To learn more about E-Verify, including your rights and responsibilities, please visit e-verify.gov

Requirements

Must-haves

Navigation & avoidance in the wild: Train policies that adapt online to cluttered 3D scenes (forests, bridges, urban canyons), complementing our stack for robust obstacle avoidance and dynamic goal-seeking.
RL-augmented planning: Fuse learned cost shaping / value functions with trajectory optimization for smooth, agile flight with tight safety envelopes and mission constraints.
Sim to Real at scale: Build scalable datasets and training loops with Isaac Lab, domain randomization, residual learning, and safety filters; validate on real drones weekly.
Human-in-the-loop shared control: Learn assistive policies that blend pilot intent, autonomy priors, and uncertainty-aware behaviors for intuitive control handoffs.
Fleet & multi-agent: Explore decentralized coordination for coverage, pursuit, and collaborative mapping with minimal comms.
Real hardware cadence: Prototype in sim, then flight test on production-ready aircraft and edge compute. Your work is meant to make a difference in the real world, not just in simulation.
Safety-first learning: We pair RL with formal constraints, estimation of uncertainty, and conservative action filters for dependable deployment in human environments.
Cross-disciplinary mentorship: Collaborate with perception, mapping, controls, and product teams to close the loop from research to field ops.
PhD student in Robotics, Machine Learning, Controls, or related field.
Strong fundamentals in RL, control theory, and motion planning; comfort with safety/robustness concepts.
Proficient in Python (PyTorch/JAX/Ray RLlib) and at least one of C++ or CUDA.
Hands-on experience with robotics simulation (Isaac Lab/MuJoCo/PyBullet) and sim2real techniques.
Experience training/deploying policies for navigation, manipulation, or locomotion on real robots or autonomous vehicles.

Nice-to-haves

Publications (CoRL, ICRA, IROS, RSS, NeurIPS).
Experience with onboard inference optimization (TensorRT, quantization, sparsity).
Familiarity with modern policy learning beyond vanilla RL: diffusion policies, IL/BC, offline RL, model-based RL.
Experience with multi-agent RL or distributed training.
Compensation for certain positions may vary based on the position's location.

Application prep

Use the role’s language and application details while tailoring your materials.

Keywords to mirror

agilemachine learningpython

Platform: ashby
Effort: quick