Paper explainers
Each paper as a short narrated video, an interactive animation of how it works, and a plain-language walkthrough with the original figures. Back to takahiromiki.com.
DAPPER: Discriminability-Aware Policy-to-Policy Preference-Based Reinforcement Learning for Query-Efficient Robot Skill Acquisition
DAPPER trains a fresh robot controller each round and asks people only questions they can actually answer — so a walking robot learns the style someone wants from far fewer answers.
Large-Scale Autonomous Gas Monitoring for Volcanic Environments: A Legged Robot on Mount Etna
Researchers put a lab-grade gas analyser on a four-legged robot and sent it across Mount Etna on its own, so that one day volcano scientists won’t have to walk into toxic fumes to take a reading.
Learning Whole-Body Humanoid Locomotion via Motion Generation and Motion Tracking
A humanoid robot learned to climb boxes, vault hurdles and take stairs the way people do — hands and knees included — by pairing an AI that sketches the next half second of human-like movement from the ground it sees with a controller that carries that sketch out safely.
Motion Priors Reimagined: Adapting Flat-Terrain Skills for Complex Quadruped Mobility
This paper teaches a four-legged robot to copy dog movement recorded on flat ground, then to learn small corrections, so the same natural gait carries it up stairs, over blocks and around obstacles to a goal.
Bridging Perception and Control for Legged Locomotion and Navigation in the Wild
Over four projects, this thesis taught a four-legged robot to use its eyes without being fooled by them, and then to use common sense, until it could hike ten kilometres down a Swiss mountain largely on its own.
Learning Quiet Walking for a Small Home Robot
Robot dogs like Sony’s aibo clatter across the floor. The researchers taught aibo, inside a simulator, to set each paw down softly, and on the real robot it walked more quietly than Sony’s own hand-tuned “quiet” mode.
Rethinking Robustness Assessment: Adversarial Attacks on Learning-based Quadrupedal Locomotion Controllers
The authors trained a second AI to make walking robots fall: it found gentle, realistic moves that topple even a DARPA-winning controller, and training against it made that controller tougher.
Learning to walk in confined spaces using 3D representation
Engineers at ETH Zurich taught a four-legged robot to see the world in 3-D and decide for itself when to duck, so it can crawl under fallen slabs and into gaps where people cannot safely go.
Learning to Open and Traverse Doors with a Legged Manipulator
A four-legged robot with an arm learned, in simulation, to open a door and walk through it without being told whether to push or pull. It works that out by feel, like a person at an unfamiliar door.
Dataset and Lessons Learned from the 2024 SaTML LLM Capture-the-Flag Competition
Can a chatbot keep a secret? Researchers turned that question into a public game: 163 teams guarded a password hidden in a chatbot’s instructions and tried to steal each other’s — and every guard was eventually broken.
Learning a Risk-Averse Locomotion Policy from Uncertainty Estimates
Researchers at ETH Zurich and TU Munich taught a four-legged robot to notice when it can’t trust what it sees, so it slows down, stops or backs away instead of walking blindly over an edge.
Learning Risk-Aware Quadrupedal Locomotion using Distributional Reinforcement Learning
The authors trained a four-legged robot to foresee the whole range of good and bad outcomes of its moves, not just the average, so that one dial can tell it, even while it walks, how much risk to take.
Identifying Terrain Physical Parameters from Vision - Towards Physical-Parameter-Aware Locomotion and Navigation
A walking robot learns to tell how slippery or soft the ground ahead is just by looking, by first learning in simulation what slippery and soft feel like, then letting its own footsteps teach its camera.
Learning robust autonomous navigation and locomotion for wheeled-legged robots
A four-legged robot with wheels for feet learned in simulation how to move and where to go, then drove itself for kilometres through Zurich and Seville.
Event-based Agile Object Catching with a Quadrupedal Robot
Researchers in Zurich gave the four-legged robot ANYmal an eye that only reports what changes, so it can spot a ball thrown at up to 15 m/s and swing a net into its path, catching 83 % of throws with every calculation done on board.
MEM: Multi-Modal Elevation Mapping for Robotics and Learning
A team from ETH Zurich and Oxford taught a robot’s map of the ground to remember not just how high the ground is, but what it looks like and what it is, fast enough to run on the robot’s own computer.
Scientific exploration of challenging planetary analog environments with a team of legged robots
Three walking robots, each with its own job, explored Moon-like and Mars-like terrain on Earth together — a sign that robots on legs could one day study the steep, loose places that wheeled rovers must avoid.
CERBERUS in the DARPA Subterranean Challenge
Four-legged robots from Team CERBERUS explored an unknown underground course for an hour, largely on their own, found 23 of 40 hidden objects and won the DARPA Subterranean Challenge. This paper explains the ideas behind that run.
Team CERBERUS Wins the DARPA Subterranean Challenge: Technical Overview and Lessons Learned
In September 2021, four walking robots, helped by a rover and guided by one person standing outside the entrance, explored a dark, smoky underground course for an hour, found 23 hidden objects and won the DARPA Subterranean Challenge. This paper explains how the whole system worked, and what the team learned the hard way.
CERBERUS: Autonomous legged and aerial robotic exploration in the Tunnel and Urban Circuits of the DARPA Subterranean Challenge
Walking and flying robots, run by one person outside, can search an unknown mine before rescuers go in. Here is how team CERBERUS built such a team and how it did in two DARPA competition rounds.
Elevation Mapping for Locomotion and Navigation using GPU
Researchers at ETH Zurich moved a walking robot’s terrain map onto its graphics chip, so it can turn hundreds of thousands of sensor points into a clean map of the ground many times a second, and they fixed the everyday glitches that make such maps lie.
Combining learning-based locomotion policy with model-based manipulation for legged mobile manipulators
A four-legged robot with an arm on its back can go where wheels can’t and do useful work there — but every swing of the arm shoves its body. This paper lets the arm’s planner warn the legs, up to 0.8 seconds ahead, how hard the arm is about to push, so the legs can brace in time.
Reconstructing occluded elevation information in terrain maps with self-supervised learning
A walking robot’s map of the ground always has blind spots behind rocks and step edges; this paper teaches a neural network to fill them in using only the robot’s own incomplete maps, by hiding extra patches and practising to guess them back.
Learning robust perceptive locomotion for quadrupedal robots in the wild
A four-legged robot learned, in simulation, when to trust its eyes and when to trust its legs, so it can walk fast over real mountains, snow and stairs without falling, even when its view of the ground is wrong.
Circus ANYmal: A Quadruped Learning Dexterous Manipulation with Its Limbs
A four-legged walking robot learned to lie on its back and turn a yoga ball with its feet, like a cat playing with a toy — the first real quadruped to handle an object this dexterously with its limbs.
Real-time optimal navigation planning using learned motion costs
A walking robot’s best route isn’t the shortest one — it’s the one its own legs handle best. This paper lets ANYmal learn those costs in simulation and plan with them on board, in about a second and a half.
Perceptive Locomotion in Rough Terrain – Online Foothold Optimization
Researchers at ETH Zurich taught the four-legged robot ANYmal to look at the ground just ahead and pick a safe spot for every foot, fast enough to keep trotting over steps, gaps and loose bricks.
UAV/UGV autonomous cooperation: UAV assists UGV to climb a cliff by attaching a tether
A drone flies a cable up a cliff and winds it around a pole, so the ground robot tied to the other end can reel itself up to places it could never drive to.
Robust Rough-Terrain Locomotion with a Quadrupedal Robot
A four-legged robot sees the ground with its own sensors, picks a safe spot for every foot and tilts its body to reach it, so it can climb steps up to 21 cm tall on terrain it has never seen, with all its computing and power on board.
Multi-agent Time-based Decision-making for the Search and Action Problem
A team of drones must find objects on a big field and carry them home before the clock runs out; this paper gives each drone one simple question to decide when to keep searching and when to cash in.
Self Localization with Active Attachment of AR Markers by Autonomous UAV
A small drone that sticks its own navigation markers onto the ceiling, so it can find its way into places nobody prepared for it.