PhyRC Challenge

We are hosting the PhyRC challenge, a competition to facilitate innovation in physical robotic caregiving. The competition has two tracks (Track 1: Fixed-base Manipulation for Robot-assisted Dressing and Track 2: Mobile Manipulation for Robot-assisted Bed Bathing), each evaluated through two phases, namely Phase 1: Simulation Phase and Phase 2: Real Robot Phase. We would like to thank Kinova for generously sponsoring a Gen 3 7-DoF robot arm for the Track 1 winning team and Hello Robot for generously sponsoring a Stretch 3 robot for the Track 2 winning team.

GitHub Documentation EvalAI

About Physical Robotic Caregiving

A report from 2021 states that over 1 billion people around the world require assistance to carry out daily tasks, and that this number will significantly increase with the aging population. Caregiving robots have the potential to reduce caregivers' workload and increase care recipients' independence. The EmPRISE Lab is hosting the PhyRC (pronounced as fai-R-C) Challenge, which stands for Physical Robotic Caregiving Challenge. With the mission of enabling robots to improve the quality of life of people with mobility limitations by assisting them with activities of daily living (ADLs), we provide realistic setups for robot-assisted dressing and bathing tasks which are important ADLs, with the hope to explore solutions to interesting rigid and deformable object manipulation and physical human-robot interaction problems in physical caregiving settings.

Competition Phases

Simulation Phase (Phase 1) - We will use RCareWorld as a simulation platform for this phase. The teams can choose which track they want to participate in (they can participate in both tracks as well) and are required to complete at least one of the caregiving tasks in the simulation environment. We will provide the simulation environments including the human avatars, robots, and environmental objects.
Real Robot Phase (Phase 2) - Select teams with the best simulation results will be invited to compete using real-world robots in May 2025 at a venue which will be announced later. Participants will be allowed to use any kind of robot they may have access to. We do not apply restrictions to the algorithms or robots. The robots need to perform the same tasks in the real world except this time, the robots will do the same tasks with a manikin (and not a real human). We will provide the real-world environment including manikins, hospital beds, wheelchairs etc. The participants will just need to ship their robots to the venue.

🥇 Prizes 🥇: The leader of Track 1 wins a Kinova Gen 3 7-DoF robot sponsored by Kinova and the winner of Track 2 gets a HelloRobot Stretch 3 robot sponsored by Hello Robot as prizes.

Task Details

Track 1: Fixed-base manipulation for robot-assisted dressing

Assistive dressing is an ADL task during which the caregiver helps the care recipient put on garments. The task setting for the competition considers using a Kinova Gen 3 7-DoF arm to help a human avatar perform the dressing task in Phase 1. For Phase 2, the participants are free to use any robots. The manikin sits in a wheelchair without handles and both arms are stretched forward. The robot needs to grasp a hospital gown, approach the wrist of the manikin, and pull the sleeve up as much as possible to dress both arms.

A demo of what the task might look like:

Teams will be scored using the following rubric:

Pick up the garment (5 pts): This task will be considered successful if the hospital gown stays in the robot's hand for 3 seconds when the robot’s hand is lifting the garment in the air.
Put one wrist in one sleeve (5 pts): The task will be considered successful if at least one of the manikin’s hands is contained within one of the gown sleeves.
Pull the gown to the other side (5 pts): The task will be considered successful if any part of the gown goes beyond the shoulder joint of the manikin, and the gown stays between the back of the chair and the manikin.
Dress the other sleeve (5 pts): The task will be considered successful if the manikin’s hand is in the other one of the sleeves.
Overall dressing (30 pts): (n%)*10+(m%)*20 (n = (length covered by the cloth)/(entire arm length of the first arm), m = same metric for the second arm).

We will also account for the efficiency by dividing the score by the time taken for the entire task. We will randomize the positions of the objects and the human configuration in the environment, and evaluate the proposed solution multiple times with randomized initial states.

Track 2: Mobile manipulation for robot-assisted bed bathing

Bed bathing is an ADL task performed during everyday nursing procedures for people with severe mobility limitations to help them maintain hygiene. The task setting for the competition considers full-body bed bathing, where a manikin is lying on the bed, and the robot needs to grab a scrubber, dip it in the water, and slide the scrubber over the entire human body while making sure that the force exerted is appropriate. The participants need to program a Stretch 3 robot to cover as much body area as possible while keeping the force within a certain threshold to make the human feel comfortable.

In Phase 1, we use the blue color to indicate the area covered by water. We will provide APIs to read the covered area over the entire top of the body and the force on the scrubber. In Phase 2, we will provide some paint so the participants can put the paint on the manikin’s body to indicate the area covered. We’ll design a scrubber with a load cell force sensor inside, and release the design files.

A demo of what the task might look like:

Teams will be scored using the following rubric:

Pick up the scrubber (5 pts): This task will be considered successful if the scrubber stays in the robot's hand for 3 seconds when the robot’s hand is lifting the scrubber in the air.
Dip the scrubber in a water tank (5 pts): This task will be considered successful if the scrubber touches the water and the water tank is not knocked over.
Move the scrubber to the manikin (5 pts): This task will be considered successful if the scrubber is in the hand of the robot, and in contact with any part of the manikin’s body, and the force does not exceed a threshold.
Perform the full-body bed bathing by sliding the scrubber over the entire body of the manikin (35 pts):
- Body coverage: (n%)*10 (n = area covered with water/entire top body)
- Force threshold: (m%)*25 (m = timestep of forces within the threshold/entire timesteps when the scrubber is in contact with the human)

Participation Guidelines

Step 0: Fill in this registration form. Check the box to join the mailing list to avoid missing any updates!
Step 1: Follow the install guide in the phy-robo-care branch to install RCareWorld. Get yourself familiarized with the codebase.
Step 2: Stay tuned for the release of the simulation environment for dressing and bathing. We will send an email to all people who stated their interest in joining the mailing list once we release the simulation environments.

Tentative Schedule

The schedule is subject to change and will be updated in a timely manner. Join the mailing list using the registration form to get direct notification.

Registration Deadline: Sep 8, 2024 (11:59pm AOE)
Phase 1 Announcement: Aug 10, 2024
Phase 1: Aug 10 - Dec 1, 2024 (4 months)
Phase 2 Announcement: Dec 7, 2024 (Top-n teams released)
Phase 2 Preparation: Dec 7, 2024 - May 2025, Teams Prep for Real-robot competition (5 months)
Phase 2 Final Preparation at Venue: May 2025, Teams practice in competition arena for a couple of days (Venue TBD)
Phase 2 Competition: May 2025, Teams compete during this period and the winners are announced at the end

Contact

If you have any questions regarding the competition, please email empriselab21 at gmail dot com with PhyRC in the subject line. If you have any questions regarding using RCareWorld, feel free to ask in our discussion forum.

FAQs

Are third party libraries allowed?
- Yes, third-party libraries are allowed. Ensure your Docker setup is configured correctly so the environment runs smoothly on the evaluation platform.
Does the evaluation platform have GPU?
- Yes, the platform supports a single GPU with 12GB of memory.
If there is a discrepancy between the score that I get locally and the score reported in EvalAI, does that mean there is a bug in the competition code?
- Differences can arise from several factors. We use a set of seeded randomizations for the tasks that are not visible to participants. Make your policy as robust as possible. Additionally, simulation speeds on local computers may differ from the evaluation platform, causing slight score variations. You are allowed multiple submissions per day, and only the highest score will be considered.
What aspects of each task are randomized?
- For bathing: the human's limb poses, the initial position of the sponge, and the initial position of the robot's base.
- For dressing: the human's arm pose.