About the Challenge
What is the BatterySwapAI 2026 Challenge?
A data science competition hosted by NORA, with data provided by Soundsensing. Participants build a system that (1) predicts the remaining useful life of replaceable batteries in IoT condition-monitoring sensors deployed across Norwegian commercial buildings, and (2) generates an efficient work plan for when and in what order to replace those batteries across the entire fleet.
Who is this challenge for?
Data scientists, engineers, and researchers interested in time-series forecasting and prognostics, decision-making and planning under uncertainty, and optimization and scheduling.
Why is this an interesting problem?
Soundsensing monitors over 1,000 air-handling units across Norway using battery-powered vibration sensors. Managing battery replacement across thousands of devices combines machine learning (degrading-asset prognostics), logistics, and operations research, and is analogous to broader predictive maintenance for assets such as pumps and ventilation fans.
Key Dates
What are the key dates?
- 13 May 2026 — Launch and registration opens
29 May 2026 (13:00–14:00) — Q&A webinar- 11 August 2026 — Registration deadline
- 14 August 2026 — Dataset release
- 21 August 2026 — Final submission deadline
- 1 September 2026 — Winner announcement
The actual competition takes place during a one-week window in August once the dataset is released.
Registration & Teams
How do I register?
Registration is open via https://nettskjema.no/a/batteryswapai2026 (also linked from the NORA challenge page). Register even if you are not yet certain you will participate — there is no penalty for registering and not submitting, and registration is how you will be notified when the dataset is released.
Can I participate as an individual, or do I need a team?
Both are allowed. Teams of up to 8 participants are welcome. If you would like to be matched with other individuals to form a team, let the organizers know and they will try to connect you with others.
Who is eligible to participate?
The competition is open to residents of Norway. Industry, academic, and mixed teams are all accepted. To qualify for prize money, participants must be Norwegian residents (a practical constraint around taxes and money transfers).
Prizes
What is the prize pool?
Total NOK 50,000:
- 1st place: up to NOK 40,000
- 2nd place: up to NOK 10,000
What does “up to” the full prize mean?
To receive the full prize, at least one team member must be a Master's student, PhD candidate, or have equivalent Norwegian academic affiliation. Teams without academic affiliation are eligible for half of the prize amount. This comes from sponsor requirements to encourage student–industry collaboration. Edge cases (postdocs, researchers, recently graduated students) will be clarified on the website before the registration deadline.
Are there conditions on the submission to qualify for prize money?
Yes — to qualify for the prize, submissions are expected to be open-sourced under the MIT license. If you prefer to keep your solution proprietary, you can choose not to submit it for the prize and retain the IP.
Dataset
What does the dataset contain?
Each sensor's time series contains two measured columns — battery voltage and temperature — plus a timestamp. Data is provided as a one-hour average over one-minute interval samples, stored as Parquet files. Location information is provided as logical IDs (building ID, room ID per battery) together with a travel-time matrix between buildings. No real-world geographic locations are shared.
How many devices and buildings are there?
- ~3–10 batteries per room
- ~1–10 rooms per building
- ~200 buildings in total (subject to split into training/evaluation)
- A few thousand devices overall
What battery and sensor are these?
The sensors are battery-powered Bluetooth Low Energy IoT devices that run repeated 1-minute measurement cycles consisting of Sleep → Sense → Send. The battery is a non-rechargeable CR2450 coin cell with Lithium Manganese Dioxide (LiMnO2) chemistry. The dataset comes from Rev 2 sensors, which have ~2 years of nominal battery lifetime (the newer Rev 3 sensors get ~4 years, but are not in this dataset). Capacity varies with temperature, production batches, and shelf life.
Do all sensors start at the same time?
No. Devices are installed at different points in time, although devices within a single building tend to be installed together. To prevent leakage, the dataset will likely be stratified or split by building, and the end-time cutoff will be variable per series — so you cannot simply infer the cutoff and use it directly as a prediction.
Is there a battery ID? Other metadata?
The time-series columns are voltage and temperature. A battery identifier is included, but it is not meaningful to use it directly as a feature — your model needs to generalize to unseen batteries, sensors, and locations in the evaluation set.
Can I use the dataset for research or publication outside the competition?
Yes. The dataset will be released under an open license for any purpose. The training data is intended to be released openly from August; the hidden evaluation set will be released later. A technical report describing the dataset is planned. The dataset is expected to be archived in the NIRD Research Data Archive (Sigma2) for citation.
Task & Evaluation
What exactly do I need to predict and produce?
Two coupled subtasks:
- Battery remaining useful life (RUL): from each device's time series, predict how much useful life remains. Modeling the uncertainty (e.g. as a distribution) may be advantageous, but the choice is yours.
- Replacement planning: using the RUL estimates together with travel times and constraints, output an ordered work plan — a data frame / CSV listing, for each day, the sequence of batteries to replace.
How is the work plan scored?
The provided evaluation function computes a cost (roughly total hours spent) given your work plan and the true end-of-life times. Cost components — split into those driven by planning/location vs. those driven by prognosis quality — include:
Influenced by planning / location:
- Time per battery change (wrench time)
- Time per room change (fixed overhead between rooms)
- Time per building change (fixed overhead between buildings)
- Travel time between locations (from the provided travel-time matrix)
- Overtime costs for working more than an 8-hour workday
Influenced by battery lifetime prognosis / time-series data:
- Penalty for replacing early — per missed hour of ideal lifetime
- Penalty for replacing late — per device-hour of downtime; this penalty is expected to be larger to reflect real impact of system failure
Final scoring weights are not yet locked down 100% and will be tuned to make the problem interesting and realistic. Details will be finalized before dataset release.
One worker per city, or multiple workers in parallel?
Not finalized. There will be a constraint on the number of workers operating concurrently and on total work hours. If the natural problem is too easy with a single worker, restrictions like part-time hour caps may be introduced to keep the planning subtask non-trivial.
Will the penalty for early vs. late replacement be the same?
Probably not. In practice downtime is more severe than premature replacement, so the cost function will reflect that asymmetry.
Are baselines provided?
Yes — example code in Python with data loading and very simple (“functional but stupid”) baselines for prognostics and planning will be released so participants can run the full pipeline end-to-end. Stronger reference baselines may be withheld to avoid biasing approaches.
How does the evaluation / leaderboard work?
You upload your code to a submission portal; it runs against the evaluation dataset. There are two splits:
- Public evaluation — drives a leaderboard updated on submission.
- Hidden evaluation — used for final scoring, not visible until results are announced.
Materials Provided
What will be released to participants?
- Training dataset (time series + location/travel-time information)
- Technical documentation describing the dataset and evaluation
- Example Python code: data loading, simple baseline prognostics, simple baseline planning
- Evaluation/scoring code
- Submission format specification
- A web portal for uploading submissions
Publication in Nordic Machine Intelligence
Can I publish a paper based on my submission?
Yes — and you're encouraged to. Participants are invited to submit a technical report to the peer-reviewed journal Nordic Machine Intelligence (NMI), which will run a special issue dedicated to BatterySwapAI. Papers are optional and are not part of the competition scoring.
Who is on the Scientific Committee?
- Fadi Al Machot — NMBU
- Mina Farmanbar — UiS (Stavanger)
- Kyandoghere Kyamakya — University of Klagenfurt, Austria
- Habib Ullah — NMBU
- Ibrahim Abdelfattah Abdelhameed — NMBU
How does the review process work?
Papers are treated as regular research articles, peer-reviewed by a list of reviewers prepared for the special issue. The issue will likely be edited by members of the scientific committee, with an editorial paper summarizing the challenge and the best results. A page limit will be set; short papers are preferred.
Is there a reference for what these papers look like?
Yes — the recently published WindAI special issue, from a previous NORA data challenge, provides a good template.
Practical Questions
Can I start working on something before the dataset is released?
Yes. You can familiarize yourself with battery RUL literature, prognostics methods, and route/scheduling optimization. A practical suggestion from the committee: look for public battery / IoT-sensor datasets and prototype your pipeline on those, so you can move quickly once the official dataset is released in August.
Will the webinar be shared?
Yes. The recording will be published on YouTube and the link shared with everyone who registered.
Who do I contact with further questions?
Kushtrim Visoka — kushtrim.visoka@nora.ai.
Organized by NORA (Norwegian AI Research Consortium). Dataset by Soundsensing. Scientific committee from NMBU, UiS, and the University of Klagenfurt. Rules and details will continue to be refined and posted on the challenge page ahead of the registration deadline.
Challenge page: https://www.nora.ai/competitions/batteryswapai/batteryswapai2026
Register: https://nettskjema.no/a/batteryswapai2026