2026 RCJ Rescue Line Rule Changes

RoboCupJunior Rescue RoboCupJunior Rescue Line
1

It’s important for us to know your opinions (each of the different teams in this sub-league), so we want to share a Draft of the New Rules for Rescue Line 2026. We’ve analyzed the various comments received during the last competition, and as a result, we’ve made some changes.

Please read this document carefully and analyze it. Feel free to share your opinions with us in the forum.

See you in Incheon, South Korea!

1. EVACUATION ZONE

1.1. LIGHTING

We propose adding lights to the evacuation zone. This will help us verify that teams working with AI should be able to work with different lighting conditions.

This rule consists of a white LED light (flashlight), mounted 10 cm high, perpendicular to the wall. It is not placed in the corners where the triangles are located. Its placement on one of the walls is arbitrary; it can be on any wall, as shown in the image.

Light
Light633×367 148 KB

In the case of teams without AI, they would have to recalibrate their light sensor; they wouldn’t be told where the light is coming from, and the calibration would have to be done on the field.

1.2. POSTS

Currently, in the rule 3.9.3, we mention that: “The evacuation zone is 120 cm by 90 cm with walls around the four sides at least 10 cm high and colored white.” It has been implicitly assumed that the posts are white.

We propose allowing the posts on the walls of the evacuation zone to be different colors; item Profiles to hold up the walls would be allowed.

This will not pose a significant challenge for teams with conventional sensors, but teams with camera systems will need to further prepare their models. This change will allow team to use the same material we already use for Rescue Maze, to have the gray aluminum profiles mounting the walls.

1.3. WALL

The walls in the rescue zone can be any color different from the white floor (except for red, green, and black, since the triangles are red and green and a dead victim is represented by black). This does not pose any difficulty for detecting victims or the triangles.

2. VICTIMS

We suggest adding fake victims in the evacuation zone. These could be printed pictures of victims on the floor in the evacuation zone or fake live victims.

Fake live victims are objects that look like real live victims (for example, silver or reflective balls), but they are not conductive and will stay inside the evacuation zone.

For each fake live victim pick up, 5 points will be deducted from the evacuation points.

3. FLOOR

In rule 3.2.3, we mention that: “Robots must be designed to navigate under tiles that form bridges over other tiles. Tiles placed above other tiles will be supported by pillars at tile corners with a square cross-section of 25mm x 25mm, making each tile entrance/exit 25 cm. The minimum height (space between the floor and the ceiling) will be 25cm.”

We propose a maximum design size of 25mm x 25mm for the pillars supporting the tiles used to create bridges over other tiles or to construct the various ramps. An example of a post for ramp construction could be one made of 20x20 mm aluminum extrusions. This does not pose any disadvantage to the teams.

4. TDP Rubric

More details on these points will be given in the rubric document.

We propose adding three sections or subsections detailing the following information, including images, flowcharts, or code snippets illustrating each part. This will help us verify that the teams using AI truly understand what they are working with and the technology they are handling.

4.1. Victim Detection.

If the team uses a camera for victim detection, explain how to develop the victim detection system, its type, and the parameters used (AI/open CV model/machine learning). If the team does not use AI but uses light sensors, describe how the system works, its parameters, and the types of sensors used.

4.2. Line-following system.

Explain how your system works using some type of AI (machine learning, OpenCV). If you don’t use AI, explain your system using light sensors.

4.3. Dataset.

Add a section to the TDP. If the robot uses AI, machine learning, etc., list them, explain how they were implemented in the robot, and include the dataset. Failure to add this information to the TDP will result in a percentage deduction or penalty (since the lack of this information demonstrates a lack of understanding of the model used by the AI, which can be interpreted as the team not understanding or creating the model).

Best,

Diego Garza Rodriguez on behalf of the 2026 RCJ Rescue Committee

3 Likes
2

Thanks for the update on the rules and for sharing the changes. Just a small note on the fake live victims. Fully agree with the idea, but if the robot picks up the victim and then release it, will that lead to deduction of points? Thanks

1 Like
3

Can I request for more details on the fake victims? If it’s a printed picture on the floor, that doesn’t sound like something that can be picked up at all. If it’s a 3D object (eg. a ball), then what’s the difference between the fake victim and the real victim? Only the conductivity?

4

Hello @axisa,

Great question! I think I know the answer but would like to get the full committee agreement before replying back, appreciate your patience, we will get back to you shortly.

Hello @Cort,

For sure! Instead of thinking what is allowed to be a fake victim, I would think it the other way around, the rules specify a proper definition of a live / dead victim, everything else organizers can add them in the effort to make fake victims as long as these objects don’t interfere with other rule definitions (for example, adding obstacles closer than 10cm to a wall and argue that is not an obstacle but a fake victim).

About your inquiry in specific, yes, there could be a printed picture in the floor to confuse camera systems, it could a 3D object that is way bigger or smaller than the definition of a victim or it could be a ball with the same size but different colour. In the one we mentioned, a gray ball that doesn’t conduct electricity would be considered a fake victim for example.

Does that help?

Thanks!

Diego Garza Rodríguez
2026 Committee

5

If the definition of a fake victim is anything that doesn’t fit the definition of a real victim, then all of the following could qualify as a fake victim…

  • 4cm sphere that is silver, reflect light, and is electrically conductive, but without an off-centered center of mass.
  • A sphere that is identical to a live victim in every way, but weights 81g.
  • A spheroid that is slightly oblate, but otherwise identical to a live victim.

I’m sure the committee intends for the differences to be more significant, but without clarity in the rules, the teams would need to be prepared to detect any differences, however small.

Perhaps I can suggest defining the fake victim like this…

Fake Victims
These are similar to the live victims, but differs in at least two of the following:

  • Shape: A fake victim is not spherical (eg. it could be a cube). If the fake victim is a spheroid, the minor diameter will be no more than 50% of the major diameter.
  • Size: A spherical fake victim will have a diameter of less than 3cm or more than 7cm.
  • Weight: A fake victim will weight no less than 120g.
  • Conductivity: A fake victim will be non-conductive.
  • Color: A fake victim will not be colored silver or grey [1].

I would also suggest that the penalty for fake victim should only apply when they are moved into the evacuation zone, and not on pick up. Properties such as conductivity can’t be easily measured without contact, and it can be difficult to determine if the victim was picked up when the ball is being probed. For example, a robot may use a claw to grip the victim and measure if it is conductive. In the process of gripping the ball, it may slightly lift it off the ground. This could be very difficult to determine. If the differing property is weight, it would be extremely challenging to determine the weight without picking up the fake victim.

[1] As a color, silver is rather vague. Depending on lighting conditions and how polished the surface is, silver may be indistinguishable from gray. If gray is to be an acceptable difference for a fake victim, then I would suggest that either…

a) The degree of “polish” for a live victim be clearly quantified. This could be quite difficult to do.
b) Both live and fake victim must be made available to the teams for pre-game calibration (…the rules do not currently guarantee this).

3 Likes
7

I fully understand how challenging it is to ensure a meaningful return on investment for teams in terms of learning value and technical challenges. Hats off to the committee for their continued efforts.

That said, I would like to offer a few suggestions and clarifications:

Clarify the use of the term “AI.”
The document should specify whether “AI” refers to Artificial Intelligence as a broad field (which encompasses a wide range of algorithms and disciplines—from deterministic state machines to expert systems, and beyond) or specifically to AI tools. I suspect the intended meaning here is the latter, and it would be helpful to state that explicitly.

Section 1.1 – Lighting
I respectfully disagree with the addition of an external light source.

Rationale 1:
It compromises fairness if external adjustments are made specifically to assist teams using AI tools rather than those developing their own recognition methods. If additional lighting is deemed necessary, its purpose should not be to advantage only AI-based solutions. Teams with robust designs typically implement their own controlled light sources, which helps maintain a fair and consistent environment.

Rationale 2:
Allowing a bright external light source from any direction could introduce unwanted noise and variability for teams relying on AI-based recognition. It would be more effective and equitable to let teams manage their own lighting conditions. External light from unpredictable directions is more likely to confuse AI systems than to improve their performance.

Section 2 – Victims
I echo what Cort mentioned in the previous post—thank you for the clarification provided there.

Section 3 – Floor
Additional clarity regarding clearance would be very helpful, as this detail is important for consistent implementation and evaluation; e.g. hallway remains ±10% of 30cm width. However, I would imagine this will apply to Maze setup too. In that 2026 draft, it specifies 28cm ±10%… This adds to confusion… Will need clarification for consistency.

Section 4.1 – Victim Detection
The phrase “…If the team does not use AI but uses light sensors…” could use clarification. Teams may employ cameras or computer vision techniques to develop their own recognition algorithms without relying on AI tools. It’s unclear why the rules would appear to favor teams using prebuilt AI tools over those developing their own algorithms. Perhaps I’m misunderstanding the intent, but further explanation would help.

Section 4.3 – Dataset
I assume the intention is for teams to upload their dataset rather than include it directly in the TDP.

– Elizabeth Mabrey

1 Like
8

Hi Dieguinilombrin, thanks for the update on rules!

Regarding 2. VICTIMS (Fake Victims)

I suggest there should be much more detailed guidelines established regarding “fake victims.”

The current proposal allows for “objects that look like real victims.” However, without strict definitions, it is unclear how similar they can be.

  • Does it include mannequins?
  • What are the size and weight limits?
  • Is it distinguishable by vision, or only by touch?

If the guidelines are too vague, it becomes a guessing game rather than a technical challenge. We strongly request a clear specification of what constitutes a “fake victim” to ensure fair play and proper preparation.

Thank you!

1 Like
9

Hello everyone!

Thank you for your continuos support and comments regarding the rule changes, the committee is closely reviewing each of team and talking about possible changes in our meetings. Will try to address your comments and concerns here.

Victim Misidentification Penalty
Thank you for bringing the concern Axisa. After reviewing this with the committee we decided to remove the penalty. It created way to many edge cases for the different type of robots that participate in the competition and the fact that Rescue Line doesn’t have a way to identify a victim separate from rescuing a victim, it makes it really challenging. We will reconsider this in the 2027 rules.

Fake Victim Definition
While we understand that having a clear fake victim definition helps teams to prepare better for the known, our goal of fake victims is to prepare teams for the unknown. We have done this in Rescue Maze for multiple years without a clear definition and it has worked really great by giving the competition organizers to adapt the fake victims based on the team’s level. For example, if running a competition where most of the teams are having difficulties rescuing a victim, most likely there will be little to no fake victims. If most teams are able to identify all the victims without problem, having more and more challenging fakes will be in place. Again, the goal of a fake is not to play teams and say “aha! You didn’t consider this pokemon figure could be a fake”, but use victim alike elements to define that. We already talked about great examples here of fake victims and would like to keep it open like that.

AI wording and definition
We used the word AI multiple times in the post because we wanted to explain the reasoning behind each of the proposed rule changes and what was on the Committee mind when proposed. We aren’t going to reference AI or similar in the rules when making the changes.

Lighting element
The goal is to have a new way of adding noise to the evacuation zone. As most of the elements, these are optional, similar to having obstacles in the evacuation zone. The goal is for teams to not rely only on vision based sensors when determining the position and location of victims and rescue zones. We are sharing this ahead of time for teams to plan for those. Some of the proposals here like adding the robot own light source and taking into consideration vision sensors position with regards to the lighting noise are great ways of tackling this change.

Thanks,

Diego Garza Rodríguez
2026 Committee

1 Like
10

Hi! In my opinion rules should give more details.

1.3 – Will all walls have the same color, or can each wall have a different one?

2 – Fake Victims

As Dieguinilombrin said, teams should define what a real victim is in order to detect a fake one. However, it would be helpful if the rules explicitly stated what the organizers are allowed to change in a victim to classify it as fake.

For example, it is not clear whether robots are expected to measure the victim’s weight to confirm it is real. From my understanding, starting this year, robots are also required to check conductivity, as otherwise they cannot confirm if a victim is real. Giving examples of fake victims with possible major changes would also work.

The rules should also clarify what exact difference (e.g., percentage) in size, weight (if weight check is required) or other properties is considered a valid indication of a fake victim.

11

Hello @rdanik,

From my understanding, starting this year, robots are also required to check conductivity, as otherwise they cannot confirm if a victim is real.

Well… in reality, for a while ago, conductivity was a requirement and part of the definition of a live victim. If a competitor organizer wanted to do this 2 or 3 years ago, it was within the rules. The only thing that we are proposing changing is making teams aware that fake victims are a possibility by explicitly calling this out in the rules, but is not a rule change that is allowing new fake victims that previously weren’t allowed. Does that make sense?

And the walls could have different color :slight_smile: We are adding this mainly so we can use white walls and gray aluminum posts to hold those walls instead of needing to paint each pole.

Thanks,

Diego Garza Rodriguez
2026 Committee

12

Hi! One quick question. If the robot fails to detect the fake victim and consider it as normall victim so in the end the robot will bring the fake victim to the red or green triangle (sorry I can’t remember what are they called :slight_smile:), will be this action penalized? If not, teams could simply program their robot to collect everything on the field and simply bring it to the “triangles”.
Thank you very much for your response

13

I must disagree with this point. The old rules specify that the evacuation zone contains precisely two live and one dead victim (rule 3.10.4). If the organizer wanted to add a fake victim (i.e. an obstacle) being neither conductive nor dead and thus not a victim, it would have to be at least 15cm tall according to rule 3.5.4. Thus teams could define an alive victim by one of its properties (within reason)

The introduction of possible fake victims into the zone would mean that the teams cannot rely on a spherical shape, color/conductivity/light-reflection (depending on the team’s way of the victims detection) and common sense anymore.
Thus, the teams need to check every property of a victim there is, to define its status accurately. I fully support Cort’s suggestion, as rules should clarify more information, otherwise the teams will have to scrutinize a victim under a microscope including checking for its center of mass to be 100% sure it’s not fake.

Regards,

2 Likes
14

Hi @rdanik, I think @Dieguinilombrin was referring to the conductivity matter, not the quantity of victims. I did a quick search and found that conductivity as a characteristic of victims has been part of the rules since at least 2015. According to the 2015 rules (section 1.8.3):

The victim represents a “person with life” and will be electrically conductive. Its surface is silver and reflects light.

I’m confident the committee will carefully address all these concerns and ensure the final rules are clear and well-defined. It’s great to see the rules evolving to keep the competition fresh and challenging. After all, encouraging innovation is what RoboCupJunior is all about!