Student Projects

Currently, the following student projects are available. Please contact the responsible supervisor and apply with your CV and transcripts.

In case you have project ideas related to any of these projects, take the opportunity and propose your own project!

We also offer Master Thesis and or exchange semesters at reknown universities around the world. Please contact us in case of interest!

Studies on Mechatronics

We offer students also to conduct their Studies on Mechatronics at our lab. In general, we recommend to do the Studies on Mechatronics in combination with the Bachelor Thesis, either as prepartory work the semester beofre or as extended study in parallel. If you want to do it independently, yiou can find prroposed projets also in the list below. Please directly apply with corresponsponding supervisor.

Traversability-Aware RL Navigation with Semantic Perception

This work explores the adaptation and evaluation of a semantics-aware reinforcement learning approach for safe robot navigation in construction environments. The existing navigation system uses a lightweight onboard segmentation model to produce pixel-level semantic labels, which are fused with depth data to provide traversability information to the navigation policy and generate velocity commands. While this approach performs well in relatively simple indoor and outdoor settings, construction sites introduce substantial challenges, including dynamic scenes with workers and heavy machinery, scattered tools, and hazardous terrain. In addition, the system must operate entirely onboard the robot, without reliance on cloud-based resources. This project will address those challenges while evaluating the adapted planner first in simulation and then deploying it in real-world construction scenarios.

Keywords

Reinforcement Learning, RL, Simulation, Robotics, Deep Learning, Real World

Labels

Semester Project , Master Thesis

Description

Most mobile robots rely primarily on geometric sensing, such as lidar or depth cameras, for navigation. While this enables basic collision avoidance, it reduces the environment to a binary distinction between “traversable” and “non-traversable.” As a result, the robot lacks semantic understanding of obstacles and cannot distinguish between objects that may be safely traversed and those that should be avoided, such as people lying on the ground or, in construction settings, tools, materials, cables, and other hazards. Construction environments pose additional challenges that go beyond those encountered in typical indoor or outdoor settings. They are highly dynamic, with workers and heavy machinery frequently entering and leaving the robot’s field of view. The scene layout changes over time as materials are moved, tools are left behind, and structures are partially assembled. Moreover, the terrain itself is often uneven, cluttered, or unstable, containing hazards such as loose gravel, protruding elements, and temporary installations like cables or hoses. These factors make reliable and safe navigation particularly difficult when relying on geometry alone. In our previous work, we incorporated semantic scene understanding into an end-to-end reinforcement learning–based navigation pipeline \cite{yang2025spatially} using a cloud-based, open-vocabulary semantic labeling service similar to \cite{Roth2024viplanner}. This enabled the robot to reason about scene content beyond geometry alone, but it relied on continuous internet connectivity and suffered from variable latency and uncertain label availability—limitations that are especially problematic in safety-critical, real-world construction deployments. This project builds on that foundation by replacing the cloud-based semantic labeling component with a lightweight, fully onboard segmentation module that predicts a fixed set of construction-relevant semantic classes in real time. By fusing these semantic predictions with depth information, the navigation policy can reason about traversability in a more informed and safety-aware manner. The objective is to enable the complete navigation stack to run entirely onboard the robot, without external connectivity, and to evaluate its performance in both simulation and real construction scenarios. Particular emphasis is placed on safety-critical behaviors, including maintaining safe distances from workers, avoiding tools and equipment, and preventing failures such as driving over cables or entering hazardous or unstable terrain.

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2026-01-23

Organization Robotic Systems Lab

Hosts Roth Pascal

Topics Information, Computing and Communication Sciences

Online Coverage Path Planning for Indoor 3D Mapping

This work investigates a coverage-oriented navigation approach for indoor 3D mapping, where mapping itself is the main planning objective. A coarse digital building model serves as prior information, and the system generates sparse waypoints that guide a robot through the space. The focus is on fast, compute-efficient waypoint generation that can run on a robot and leverage an existing navigation stack. The method is evaluated in simulation, with optional testing on a real robot.

Keywords

Robotics, Navigation, Path Planning, Learning

Labels

Semester Project , Master Thesis

PLEASE LOG IN TO SEE DESCRIPTION

More information

Open this project...

Published since: 2026-01-23

Organization Robotic Systems Lab

Hosts Richter Julia

Topics Information, Computing and Communication Sciences

Behavioral Foundation Model with test-time adaption for forceful tasks

We aim to develop forceful-BFM to enable whole-body control with forceful interactions in a promptable way, and achieve test-time adaptation for interactions under partial observation.

Labels

Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2026-01-21 , Earliest start: 2026-01-21

Applications limited to ETH Zurich , EPFL - Ecole Polytechnique Fédérale de Lausanne

Organization Robotic Systems Lab

Hosts Zhang Chong

Topics Engineering and Technology

Generalized Neural Mapping

We want to push forward generalized neural mapping methods for legged robots.

Labels

Semester Project , Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2026-01-21 , Earliest start: 2026-01-22

Organization Robotic Systems Lab

Hosts Zhang Chong

Topics Engineering and Technology

Sim2Real with UMV Robot using Meta RL

Reinforcement learning is a powerful tool for synthesizing robust locomotion strategies for both legged and wheeled robots. However, successful real-world deployment typically relies on extensive domain randomization and careful reward shaping, which are often time-consuming and difficult to generalize. In this project, we investigate meta–reinforcement learning as a potential alternative. The proposed approach aims to fine-tune (on-policy and on-line) policies trained in simulation using limited real-world data. The ultimate goal is to enable reliable real-world deployment of highly agile parkour skills on bike-like robotic platforms.

Keywords

sim2real, meta RL, continue learning, live-long learning

Labels

Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2026-01-20

Organization Robotic Systems Lab

Hosts Jenelten Fabian

Topics Engineering and Technology

Perceptive Arm Motion Planning and Control for Heavy Construction Machine Tasks

In this work we would utilize reinforcement learning, neural network actuator modeling, and perception for the control and arm motion planning of a 40ton excavator with a free-swinging gripper. The project will be in collaboration with Gravis Robotics, ETH spinoff working on the automation of heavy machinery.

Keywords

reinforcement learning, perception, hydraulics, excavator, manipulation, industry

Labels

Semester Project , Collaboration , Master Thesis , ETH Zurich (ETHZ)

Description

This project aims to develop an advanced autonomous solution for material handling tasks performed by large-scale construction machinery, exemplified by the 75-ton Liebherr LH80 excavator. A primary application focus is the complex operation of loading and unloading ships, which demands highly precise movements to avoid collisions with surrounding obstacles. The inherent challenges include managing the complex dynamics of hydraulic actuators and the free-swinging end-effector joint, tasks currently achievable only by highly skilled human operators. Building upon previous research [1], this work will establish a comprehensive pipeline for 3D control and collision-free motion planning of the excavator's arm and cabin. Due to the difficulties in accurately modeling hydraulic systems, reinforcement learning (RL) algorithms are employed for joint control. To minimize unwanted swinging of the end-effector during operations, both high-level and low-level control strategies are learned within a single RL policy. A key focus of this project is the integration of perceptive information into the RL framework. This involves incorporating 3D sensing data from LiDAR scans and semantic segmentation from camera images. This rich sensory input is crucial for detecting and classifying the material being handled and ensuring safe operation in dynamic environments with moving objects. The ultimate goal is to replace existing sampling-based motion planners with a unified RL agent. This agent will be capable of perceptive planning and control, adeptly managing kinodynamic constraints and navigating arbitrary-shaped obstacles in real-time to achieve efficient and feasible motion plans. The project will be developed in collaboration with Gravis Robotics, an ETH spinoff from the Robotic Systems Lab (RSL) working on the automation of heavy machinery (https://gravisrobotics.com/). [1] Spinelli, F.A., Egli, P., Nubert, J., Nan, F., Bleumer, T., Goegler, P., Brockes, S., Hofmann, F. and Hutter, M., 2024. Reinforcement learning control for autonomous hydraulic material handling machines with underactuated tools.

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2026-01-18 , Earliest start: 2025-07-07

Organization Robotic Systems Lab

Hosts Egli Pascal Arturo , Terenzi Lorenzo , Spinelli Filippo , Spinelli Filippo

Topics Information, Computing and Communication Sciences , Engineering and Technology

Bimanual Dexterity: Coordinated Dual-Arm Policies for Dexterous Object Manipulation

Dexterous bimanual manipulation remains a difficult challenge, requiring tight coordination between both arms and hands—especially when handling large objects. While humans perform such tasks effortlessly, robotic systems often lack the precision and adaptability needed. This project aims to tackle this problem by learning coordinated dual-arm and hand policies from a few human demonstrations, followed by reinforcement learning in simulation for real-world deployment.

Keywords

Dexterous Manipulation, Bimanual Manipulation, Learning from Demonstrations, Reinforcement Learning, Humanoid Robotics

Labels

Semester Project , Master Thesis

Description

Bimanual manipulation is key for enabling robots to perform complex real-world tasks such as handling large or bulky objects. Humans naturally coordinate both arms and fingers with fine-grained dexterity, a capability that remains challenging for robotic systems. This project aims to bridge that gap by enabling robots with dual arms and dexterous hands to manipulate large objects through coordinated control. Recent progress in reinforcement learning (RL) and sim-to-real transfer has enabled impressive results in dexterous in-hand manipulation [1]. However, bimanual dexterous manipulation remains relatively underexplored. While foundation models trained on large-scale dual-arm datasets [2] offer generality, they often fall short in terms of precision and adaptability—key strengths of RL-based methods. Another direction involves using general-purpose motion trackers pretrained on broad datasets and subsequently fine-tuning them for interaction control using RL [3]. In this project, you will extend the current bimanual manipulation capabilities for large-object reorientation tasks, where success critically depends on the coordinated use of both arms—due to object size, weight, or required manipulation complexity. Rather than relying on large pretrained models, we propose to bootstrap learning using just one or a few human demonstrations, inspired by recent successes in demonstration-guided RL for highly articulated systems [4]. We will then employ RL in simulation to train robust, coordinated dual-arm controllers, with the final objective of deploying these policies on real dual-arm robotic platforms for complex, real-world tasks. **References** [1] Handa, A., Allshire, A., Makoviychuk, et. al. (2022). DeXtreme: Transfer of Agile In-Hand Manipulation from Simulation to Reality. arXiv. [2] Liu, S., Wu, L., Li, B., et. al. (2024). RDT-1B: A Diffusion Foundation Model for Bimanual Manipulation. arXiv preprint arXiv:2410.07864. [3] Li, K., Li, P., Liu, T., et. al. (2025). ManipTrans: Efficient dexterous bimanual manipulation transfer via residual learning. arXiv.org. [4] Sleiman, J. P., Mittal, M., & Hutter, M. (2025). Guided Reinforcement Learning for Robust Multi-Contact Loco-Manipulation. ICRA.

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2026-01-13 , Earliest start: 2026-01-31 , Latest end: 2026-08-31

Organization Robotic Systems Lab

Hosts He Junzhe , Mittal Mayank , Bhardwaj Arjun

Topics Information, Computing and Communication Sciences

PerceptivePush: Object pushing for a quadrupedal manipulator with onboard perception

This project will tackle the problem of training a control policy that can move and re-orient an unknown object using only onboard perception.

Keywords

End-to-end RL, Perceptive Manipulation, Mobile Manipulation, Non-Prehensile Manipulation

Labels

Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2026-01-06 , Earliest start: 2026-02-01 , Latest end: 2026-10-31

Organization Robotic Systems Lab

Hosts Bhardwaj Arjun , Dadiotis Ioannis

Topics Information, Computing and Communication Sciences , Engineering and Technology

Vision-Based Robot Learning for Excavation Tasks: From Lab-Scale Manipulators to Heavy Machinery

This project investigates autonomous control of heavy machinery through visual learning, using a lab-scale robot manipulator in a sand environment. We explore online learning approaches—including VLM-based reward learning, model-based RL (TD-MPC2), diffusion-based policies, and hybrid methods—to develop control policies for precise reaching and granular material manipulation tasks such as digging and scooping, relying solely on stereo camera observations. The long-term goal is to transfer successful methods to full-scale excavator control, improving safety and efficiency in construction.

Labels

Semester Project , Master Thesis

PLEASE LOG IN TO SEE DESCRIPTION

More information

Open this project...

Published since: 2026-01-02 , Earliest start: 2026-01-01 , Latest end: 2026-12-31

Applications limited to ETH Zurich , EPFL - Ecole Polytechnique Fédérale de Lausanne

Organization Robotic Systems Lab

Hosts Canales Claudio , Nan Fang

Topics Information, Computing and Communication Sciences , Engineering and Technology

Gaussian Splatting Super Resolution

Radiance fields like 3D Gaussian Splatting enable fast, high-fidelity scene reconstruction from real-time data. However, reconstruction quality improves in noticeable stages—e.g., coarse at 500 epochs, detailed at 5000. This project explores whether AI-based upscaling and denoising techniques can be used to predict and accelerate these improvements during training. By anticipating reconstruction quality transitions, we aim to reduce training time while maintaining visual fidelity—advancing the efficiency of radiance field learning.

Keywords

Gaussian Splatting, Computer Vision, Reconstruction

Labels

Master Thesis

Description

Work Packages

Requirements

More information

Open this project...

Published since: 2025-12-09 , Earliest start: 2025-11-01 , Latest end: 2026-07-31

Applications limited to ETH Zurich

Organization Robotic Systems Lab

Hosts Patil Vaishakh , Wilder-Smith Max

Topics Information, Computing and Communication Sciences

Walking Through Fog with Radar Splatting and RL

Help push the boundaries of robotic perception in foggy and visually degraded environments. This project focuses on enhancing the navigation capabilities of the quadruped robot Anymal using radar to perceive through fog. You'll simulate challenging conditions in NVIDIA IsaacLab and IsaacSim, generate clear 3D maps using 3D Gaussian Splatting, and train reinforcement learning-based navigation policies for real-world deployment. Combining advanced sensor fusion with cutting-edge RL, this project offers a hands-on opportunity to explore next-gen robotic vision and autonomy.

Keywords

Gaussian Splatting, Robotics, RL,

Labels

Master Thesis

Description

Work Packages

Requirements

More information

Open this project...

Published since: 2025-12-09 , Earliest start: 2025-11-01 , Latest end: 2026-07-31

Applications limited to ETH Zurich

Organization Robotic Systems Lab

Hosts Wilder-Smith Max , Patil Vaishakh , Roth Pascal

Topics Information, Computing and Communication Sciences

Propose Your Own Robotics Project

This project invites you to step into the role of an innovator, encouraging you to identify challenges you are passionate about within the field of robotics. Rather than working on predefined problems, you will have the freedom to propose your own project ideas, address real-world issues, or explore cutting-edge topics. This project allows you to define your own research journey.

Keywords

Robotics, Research

Labels

Semester Project , Bachelor Thesis , Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-12-09 , Earliest start: 2025-01-27

Organization Robotic Systems Lab

Hosts Schwarke Clemens , Bjelonic Filip , Klemm Victor

Topics Information, Computing and Communication Sciences

Multistable robotic tensegrities for dynamic shape-shifting

This thesis will combine robotic hardware prototyping with rigorous first principles-based modeling. The goal: tap into tensegrities’ nonlinear properties, such as multistability and instability, to realize robotic modules (bodies, arms, legs, etc.) that form the basis of dynamic shape-shifting systems.

Labels

Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-12-04 , Earliest start: 2026-01-01 , Latest end: 2026-08-01

Organization Robotic Systems Lab

Hosts Baines Robert

Topics Information, Computing and Communication Sciences , Engineering and Technology

Generative Environment Design: Using LLMs to bridge Sim-to-Real Gap for Locomotion

This project proposes an LLM-driven framework to automate the design of complex simulation environments for robust legged locomotion. By translating natural language descriptions directly into low-level physics parameters, we replace tedious manual tuning with semantic reasoning. The study focuses on training adaptive policies for quadrupedal or wheeled-legged robots, aiming to achieve robust performance across diverse, language-generated terrains.

Keywords

Large Language Models, Locomotion, Sim-to-Real, Domain Randomization, Physics Simulation, Reinforcement Learning.

Labels

Semester Project , Bachelor Thesis , Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-12-03

Organization Robotic Systems Lab

Hosts Palma Emilio , Roth Pascal

Topics Mathematical Sciences , Information, Computing and Communication Sciences , Engineering and Technology , Physics

Reinforcement Learning for Excavation Planning In Terra

We aim to develop a reinforcement learning-based global excavation planner that can plan for the long term and execute a wide range of excavation geometries. The system will be deployed on our legged excavator.

Keywords

Keywords: Reinforcement learning, task planning

Labels

Semester Project , Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-12-01 , Earliest start: 2026-02-02 , Latest end: 2026-09-30

Organization Robotic Systems Lab

Hosts Terenzi Lorenzo

Topics Information, Computing and Communication Sciences

Foundation models for generalizable construction machine automation

Autonomous operation of heavy construction machinery promises major gains in safety, efficiency, and scalability for the multi-trillion-dollar construction industry. Excavators present a unique challenge, combining the navigation demands of autonomous driving with the precision of robotic manipulation in unstructured, dynamic environments. This project explores Vision-Language-Action (VLA) and diffusion-based policy learning to develop generalizable excavation policies. Leveraging large-scale demonstration data naturally collected during human operation, we aim to train multimodal control models capable of adapting across sites, lighting conditions, and machine types. The approach will be evaluated through simulation benchmarks and, if possible, real-world trials, advancing the path toward scalable, generalist autonomy for robotic construction.

Labels

Semester Project , Master Thesis

PLEASE LOG IN TO SEE DESCRIPTION

More information

Open this project...

Published since: 2025-12-01 , Earliest start: 2025-08-31 , Latest end: 2026-03-31

Applications limited to ETH Zurich , EPFL - Ecole Polytechnique Fédérale de Lausanne

Organization Robotic Systems Lab

Hosts Nan Fang , Zhang Weixuan

Topics Engineering and Technology

Perceptive Generalist Excavator Transformer

We want to develop a generalist digging agent that is able to do multiple tasks, such as digging and moving loose soil, and/or control multiple excavators. We plan to use a decoder only GPT model, trained on offline data and potentially online RL as posttraining, to accomplish these tasks.

Keywords

Offline reinforcement learning, transformers, autonomous excavation

Labels

Semester Project , Master Thesis

PLEASE LOG IN TO SEE DESCRIPTION

More information

Open this project...

Published since: 2025-12-01 , Earliest start: 2026-02-02 , Latest end: 2026-10-31

Organization Robotic Systems Lab

Hosts Werner Lennart , Terenzi Lorenzo , Nan Fang

Topics Information, Computing and Communication Sciences

Reinforcement Learning for Particle-Based Excavation in Isaaclab / Newton

We want to train RL agents on our new particle simulator, accelerated on the GPU via warp in Isaaclab and Newton.

Keywords

particle simulation, omniverse, warp, reinforcement learning

Labels

Semester Project , Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-12-01 , Earliest start: 2026-02-01 , Latest end: 2026-07-31

Organization Robotic Systems Lab

Hosts Terenzi Lorenzo

Topics Information, Computing and Communication Sciences

Self-Supervised LiDAR–RGB Encoders for Autonomous Earthworks

This project adapts recent self-supervised 3D and 2D–3D representation learning methods (Sonata/Concerto-style encoders) to build a robust LiDAR–RGB backbone for earthmoving robots. The goal is a reusable encoder that powers perception for excavation, loading, traversability, and safety on real construction sites, and runs in real time on our robots.

Keywords

LiDAR–RGB fusion, Self-supervised learning, 2D–3D representations, Earthmoving robotics, Perception

Labels

Semester Project , Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-12-01 , Earliest start: 2026-02-02 , Latest end: 2026-08-31

Organization Robotic Systems Lab

Hosts Terenzi Lorenzo

Topics Engineering and Technology

RL Finetuning for Generalized Quadruped Locomotion

This project investigates the potential of reinforcement learning (RL) fine-tuning to develop a single, universal locomotion policy for quadruped robots. Building on prior work in multi-terrain skill synthesis [1], we will probe the limits of generalization by systematically fine-tuning on an ever-expanding set of diverse environments. This incremental approach will test the hypothesis that a controller can learn to robustly navigate a vast range of terrains. As a potential extension, procedural terrain generation may be used to automatically create novel challenges, pushing the boundaries of policy robustness.

Keywords

Reinforcement Learning, Quadruped Locomotion

Labels

Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-11-28 , Earliest start: 2025-06-15

Organization Robotic Systems Lab

Hosts Schwarke Clemens , He Junzhe

Topics Information, Computing and Communication Sciences

Close-Proximity Human-Aware Locomotion

In order to safely navigate a quadrupedal robot through a dense crowd, two conditions need to be satisfied: (1) The robot body does not collide with any of the people in the scene, and (2) the robot does not step on anyone’s feet. The goal of this thesis is to train a locomotion policy that can safely traverse environments with dynamically moving bodies. The problem is treated as an inverse stepping-stone problem, using binary segmentation of the human (foot) areas to create forbidden zones within the elevation map.

Keywords

Robotics, Reinforcement Learning, Crowd Navigation, Human-Robot Interaction

Labels

Master Thesis

PLEASE LOG IN TO SEE DESCRIPTION

More information

Open this project...

Published since: 2025-11-27 , Earliest start: 2026-02-01 , Latest end: 2026-10-31

Organization Robotic Systems Lab

Hosts Scheidemann Carmen

Topics Information, Computing and Communication Sciences

Free-Viewpoint Teleoperation via Dynamic Gaussian Splatting

Immersive teleoperation should feel natural and precise—like reaching in with your own hands. Most prior work shows this only in static scenes, where the operator guides a robot to an object without truly changing the world [1]. Real tasks aren’t that simple: the moment you grasp, move, or reorient something, the scene changes. If the 3D view doesn’t update, depth cues fade, confidence drops, and mistakes creep in. We propose a single-arm teleoperation system that keeps the 3D world live. Using 3D Gaussian Splatting, the scene is continuously refreshed as objects move, so what you see in VR matches what the robot sees right now. With just a wrist camera and small, smart arm adjustments to avoid occlusions (no extra camera arm), the operator keeps a clear view of the action. In VR, you can pick helpful viewpoints on the fly, working with a photorealistic, always-current reconstruction that boosts precision, speed, and confidence. Moreover, you will get hands-on time with the real robot (you really do get to play with it), with proper safety and supervision. Strong results may also lead to a conference or journal paper submission at the end of the project.

Keywords

Dynamic Gaussian Splatting, Teleoperation, Robotics

Labels

Semester Project , Master Thesis

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-11-25 , Earliest start: 2025-12-01 , Latest end: 2026-06-01

Applications limited to ETH Zurich

Organization Robotic Systems Lab

Hosts Wilder-Smith Max , Wang Shengzhi , Patil Vaishakh

Topics Information, Computing and Communication Sciences , Engineering and Technology

Push or pull? Unifying mobile manipulation skills for a quadrupedal manipulator

This thesis will focus on unifying multiple different mobile manipulation skills into a single controller for mobile manipulation with a quadrupedal manipulator

Keywords

mobile manipulation, robotics, control, reinforcement learning, quadrupedal manipulation

Labels

Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-11-14 , Earliest start: 2026-01-07

Organization Robotic Systems Lab

Hosts An Tianxu , Dadiotis Ioannis

Topics Information, Computing and Communication Sciences , Engineering and Technology

Spatially-Enhanced Linear-Attention Transformer for End-to-End Navigation

Transformer architectures have proven highly effective in sequence modelling, but their standard self-attention mechanism suffers from quadratic complexity in sequence length. The work by Transformers are RNNs: Fast Autoregressive Transformers with Linear Attention (Katharopoulos et al., 2020) shows how self-attention can be reformulated as kernel feature maps with a linear complexity implementation. Separately, in navigation and robotics, end-to-end learning of navigation policies (e.g., via reinforcement learning) has shown strong promise; however, many architectures (such as standard RNNs) struggle to build spatial memory — that is, integrating observations from different viewpoints into a coherent spatial representation. For example, the recent work Spatially‑Enhanced Recurrent Memory for Long‑Range Mapless Navigation via End‑to‑End Reinforcement Learning (Yang et al., 2025) proposes Spatially-Enhanced Recurrent Units (SRUs) to address this. This motivates the question: can we combine the efficiency and scalability of linear-attention transformers with a spatial-enhancement mechanism tailored for navigation, thereby achieving end-to-end navigation with long-term spatial memory and efficient inference?

Labels

Semester Project , Master Thesis , ETH Zurich (ETHZ)

PLEASE LOG IN TO SEE DESCRIPTION

More information

Open this project...

Published since: 2025-10-28 , Earliest start: 2025-11-15 , Latest end: 2025-12-15

Organization Robotic Systems Lab

Hosts Yang Fan

Topics Information, Computing and Communication Sciences

Student Theses at RAI

The Robotic Systems Lab offers a number of student projects at RAI or in collaboration with RAI. Please check the individual projects and directly apply to the supervisors.

Note on plagiarism

We would like to suggest every student, irrespective of the type of project (Bachelor, Semester, Master, ...), to make himself/herself familiar with ETH rules regarding plagiarism