Robotics Master Thesis; EPFL

Robotics Master Thesis (EPFL/Embodied AI)

Join to apply for the Robotics Master Thesis (EPFL/Embodied AI) role at Embodied AI

We show four master thesis projects for the coming semester, both as an academic project (collaboration with EPFL CREATE Lab) and led solely by Embodied AI.

Simulated environments serve as a backbone for robot learning, testing controllers, and robot Sim-to-real of a tendon‑driven compliant humanoid development without the need for real hardware. While traditional robots are commonplace in simulation, novel hardware is not yet standardized in simulated environments.

In this project, you will work on developing a simulation of a tendon‑driven compliant humanoid robot using the Isaac Sim environment. Starting from the robotic arms, you will gradually expand towards a full mobile‑robot (N.B.:
The humanoid here refers to a torso connected to a wheeled base, so legged locomotion is not considered in this project).

• Development of a simulated tendon‑driven robotic arm
• Development of a simulated full‑humanoid
• Sim‑to‑real matching with existing hardware
• Compliant interactions modeled in simulation

• Using Isaac Sim (preferably on complex robots)
• Developing simulated robots
• Software engineering tools (Linux, Git, Python)
• Familiarity with physical hardware

(Planned as a joint project with the CREATE Lab and Embodied AI, however conversion into an industry‑led thesis is possible upon discussion.)

As capable humanoid robots become omnipresent, the ability to effectively operate one is a critical step towards putting such systems into performing useful tasks. Teleoperation serves as a key step towards immediate operation of such robots alongside a tool for data collection for learning‑based controllers.

In this project, you will explore and develop teleoperation interfaces for a lightweight, tendon‑driven humanoid robot. Starting from a leader‑follower arm design, we will explore the possibilities of other control inputs and feedback mechanisms that can be implemented.

• Leader‑arm style control for a humanoid (two arms, head, torso, wheels base)
• Experiment on teleoperation interfaces (motion capture, velocity input, joysticks, etc)
• Experiment on feedback systems (haptic feedback, vibration, visual, etc)
• Benchmarking and quantitative analysis

• Working with real robots (prototyped and purchased)
• Mechatronics prototyping (integrating 3D prints, electronics boards, Raspberry Pi, etc)

(Planned as a joint project with the CREATE Lab and Embodied AI, however conversion into an industry‑led thesis is possible upon discussion.)

Learning‑based controllers (specifically those based on imitation learning) present a promising direction towards autonomous applications in real world situations. Despite these advances, robots struggle on a number of fronts: long‑horizon tasks, recovery, large task adaptations, etc.

For practical deployment, an intermediate step is considered human‑supervised semi‑autonomy, where a robot is able to execute small segments of autonomous tasks, which are coordinated by a human. In this project, you will explore how to blend cutting‑edge learning algorithms with teleoperation.

• Explore state‑of‑the‑art learning‑based algorithm on real robots
• Implement and experiment on different transition methods between autonomous and teleoperated motion

• Working with real robots (prototyped and purchased)
• Implementation of learning‑based controllers
• Software and embedded hardware interfacing

(Planned as a joint project with the CREATE Lab and Embodied AI, however conversion into an industry‑led thesis is possible upon discussion.)

The head of a humanoid has a unique dual function of practicality and aesthetics/feel. Cameras and other interfaces must be controlled for the robot’s operation, while the subtle motion of the head should communicate its intentions to surrounding humans through its motion (or…