Parkour Challenge

An ambitious Goal

We all dream of having more time: time to spend traveling and exploring, time to read and learn, time to teach our children and have our children teach us, and we dream of time out – from the mundane, boring tasks that prevent us from becoming who we are truly meant to be as human. The task of robotics is to enable all of this to be possible, but progress in robotics is slow – robots remain expensive and one promising robot startup after the next dies simply because the best we can build does not provide enough value – robots are just not capable enough. Or too slow… so how about making robots more daring?

We believe that just as humans, robots grow with the challenge they’re facing – so it’s time for a new challenge. A challenge in the spirit of embodied intelligence, one that we humans easily cope with, but robots fail – and static stability is not an option:

The Roboy Parkour Challenge

Approaching a vision with a challenge

Humanoids exhibit hugely complex and dynamic interaction with unstructured environments.

Biological musculoskeletal systems show locomotion performance in unstructured and complex environments which is still unparalleled and unmatched by any robotics solution today.

However, recent developments have shown promising new robots which exhibited dynamic locomotion in complex environments but are limited to very few labs worldwide.

To find broad application of such robots outside of pure research, they need to become more robust, cheaper and widely available.

In the past, international science competitions have shown a great potential to unify and compare performance and thereby accelerate research in the respective fields towards specific goals. Notable examples are the DARPA grand challenge in the early 2000s which kick started autonomous driving, the DARPA robotics challenge, the ongoing series of RoboCup challenges and numerous computer vision and AI challenges.

Boston Dynamic’s ATLAS showing complex bipedal locomotion including jumps in a parkour like challenge.
MIT’s Mini Cheetah showing unparalleled dynamic performance in quadrupeds including backflips.


A central approach to build robots that interact with the real world is called embodiment. Instead of the classical approach of building rigid robots and then finding controllers that can handle the locomotion in the given environment, the approach of embodiment is to co-evolve the robots electromechanical system with the controller. This allows to offload tasks such as vibration dampening or shock absorption to material properties of the robot instead of requiring very high bandwidth control.

To date however, there are no challenges that explicitly lie the focus on this principle and it thus remains under-explored. We do believe however, inspired by the performance of biological musculoskeletal systems, that embodiment is a promising pathway towards agile robot locomotion in unstructured environments.

Anatomy of a good robotics challenge


  • Robots break easily
  • The environment should prevent statically stable approaches to solving the problem
  • The environment should be non-trivial to navigate
  • The environment should be replicable
  • Challenging but achievable


  • Create a soft environment, that prevents hard impacts
  • Use a soft environment, where staticially stable is not an option
  • Use an environment with multiple obstacles including a climbing section.
  • Create a standardized environment, which can be easily replicated by any interested party.
  • Work with the community to define the challenge roadmap


Vision: Beat the fastest human on an inflatable parkour race track!

Approach: Establish base-line, start with a minimally solvable solution (any robot, in simulation, hard track), make harder (legged robots, real track, soft track), repeat until vision achieved.

Challenge structure

Round 0 – 7.9.2019 and ongoing – Establishing Baseline

If you compete, you want to know what you are up against! Hence starting from the 7.9.2019, we are recording numerous runs from humans, to understand how fast we need to become eventually.

Unfortunately the weather conditions prevented us from doing proper first assessment, the fastest run so far was 8s – we are very positive that before any robot gets close to this time, we will have ample time to collect enough data.

Slow-motion of complex human interaction with the soft surface
Second perspective on first run

Round 1 – 19.9. – 8.12. – Simulation Challenge

WP 3D Thingviewer need Javascript to work.
Please activate and reload the page.

Given the 3D model of the track, build a simulated robot that gets through the track fastest.

The best Submission wins 1000€

The “robots” in this simulation challenge comprise of an arbitrary combination of rigid shapes, with arbitrary geometry, connected by rotary or ball-in-socket joints.

The robot and the simulation needs to satisfy to constraints on the right.

Participants can register for the challenge by joining the Telegram-Channel:

Submission: Submit a Gazebo SDF and all the code necessary to run the simulation on a vanilla Gazebo 10.0 installation to (We know this is very preliminary, we are currently working on a challenge platform in collaboration with Mindfire Solve and will switch the challenge to it once it is mature enough.)

Scoring: The summed time of the best 8 out of 10 runs. Runs where the robot falls off the track are counted as equivalent to the slowest run + 10s per run.

The rules are expected to evolve in the first weeks as we are finding loopholes.

  • Max 9 rigid bodies
  • Only rotary and ball-in-socket joints
  • No self-collision between the robots bodies
  • Minimum total mass: 5kg
  • Minimum mass per rigid body: 0.1kg
  • The Robot needs to fit into a box of 1 x 1 x 1m when it starts
  • Maximum torque per rotational actuator: 2Nm
  • Initial position on the track: COM is initialized randomly in an area of 0.5 x 0.5m at the beginning of the track.
  • A run is successful, when the robot touches the finish line, without falling off the track
  • Assume full knowledge of the robots position and the world’s properties
  • The controller needs to run real-time on a modern Laptop. You may submit a pretrained controller though.


Round 2 – 6.12. – 8.12. – Hack the Track

1 Weekend, 1 Track, 1 Hackathon and a surprise theme fitting the occasion. In a hackathon we’re approaching the physical race track’s soft properties in collaboration with a top secret program soon to be officially announced by TUM Chair of Entrepreneurship and partners. But as always, performance is king and the robot that gets furthest wins!

The winner gets 500€.

The details of the challenge will be made available at the beginning of the challenge itself.

Round 3+ – to be defined

Given the complex nature of this challenge it is hard to predict the outcome of the first two rounds and the best path forward. That is why we will be using the results from the first 2 rounds and the baseline data to come up with a comprehensive challenge roadmap that will then define the next steps.


The initial runs are targeted at interested researchers and labs in order to create a baseline interest. Later on we hope to build a community around these challenges whether more advanced hardware challenges for research groups and larger teams, whereas simulation challenges can be completed by individual researchers, student teams or even as part of class coursework.

We have a Telegram channel, which we use to discuss the competition:

Since the challenge will be comprised of a number of repeating smaller challenges the metric will be different for each round. Please check with their respective found that some going right now or upcoming to see how you can win.

We will make available all the data that record under CC-BY 4.0 license. Data depicting persons will either be anonymized or will have the necessary paperwork done to allow for release.

Most importantly, it’s something where we have no experience in robotics with, but humans are very good at. The traditional way of doing robotics is to reduce complexity as much as possible – creating so called lab conditions – and then trying to slowly expand to increasingly complex environments.

We believe that one underexplored variable is robot morphology. (I.e. changing properties of the robot bodies, including its morphology, kinematics, actuation, …), and those naturally cannot be explored in lab conditions, as the result is a function of the environment. A soft parkour track is the ideal middle-ground between lab conditions and the real world. Every team can simply buy one for an affordable price and thus has identical conditions. But the soft nature makes the task at hand variable and prohibits too simple solutions!

Tracks are being produced by Viva Inflatables BV in the Netherlands. You can either contact them directly or we can handle the logistics and deliver a track to your door for 6999€ + VAT. If you are interested in obtaining one, please drop us an email at:

Currently, we provide a simplified rigid surface model and an accompanying SDF.

Live events

7.9.2019 – Roboy Finals

19. + 20.9.2019 – IEEE Conference of Cyborgs & Bionic Systems, Munich

During the IEEE Conference of Cyborgs and Bionic Systems, we officially announced the competition to the public.

Scientific Advisory Board

The board will be announced shortly.

Interested to shape this collaboration from a science perspective to help us achieve the goal? Write to, we’ll be happy to have you involved!

Collaboration Partners

This competition is in collaboration with the upcoming collaborative science & problem solving platform of Mindfire:

Interested in supporting the challenge? We are actively looking for supporters that share our vision.

Most importantly:
– If you have robotics hardware with excellent simulation models?
– the possibility to create a soft simulation model of the race track
– In kind or cash prices for future rounds of the competition

Talk to us: