Research
Adaptive robotics education and digital twin research
HCR is an academic research instrument for Blockly programming assessment, adaptive learning, and digital-twin robotics education. Participants choose whether to join before the client creates a player or session.
Participation
Choose whether to participate before the study starts
The client creates no simulator session or player identifier until a person actively chooses to participate in this academic study.
Required study response
Compiled Blockly Program IR, challenge version, score, and technical session or submission identifiers required for replay and analysis.
Optional context
Primary language as a coarse code such as zh or en, and current UTC offset in minutes. The primary Participate action enables both; More settings can disable either one.
Not intentionally collected
No browser fingerprint, precise IANA time-zone name, inferred location, advertising identifier, or cross-site tracking profile.
How we describe the data
We do not call the data fully anonymous. Study records may retain session identifiers, and the server keeps routine security logs, so we describe the data as de-identified.
Framing
Two problems, one system
Pedagogical
A class moves at one speed
Hardware costs and supervision limit robotics teaching, so a class often receives one fixed sequence. Students do not learn at one fixed pace, and the mismatch affects both ends of the class.
Measurement
Programming ability is hard to score
Unlike a multiple-choice answer, a program has an open-ended response space and a continuous outcome. Standard adaptive-testing methods assume neither, so they must be adapted before they can measure programming performance.
Contributions
What is new here
- 01
Adaptive selection over a generated item bank
Challenges do not come from a fixed list. Item families generate and calibrate candidates, then Fisher information selects one near the learner’s current ability estimate. The bank can grow without someone authoring every level by hand.
- 02
Guaranteed-solvable generated items
Procedural generation and reachability are in tension: a plausible-looking target may be unreachable. Solving each candidate before it is served converts that from a hope into a precondition.
- 03
Continuous scores in a dichotomous estimator
Programming performance is continuous; the estimator available is not. An order-preserving remap around a per-item mastery threshold preserves the ordering while keeping the raw score intact for analysis.
- 04
Determinism as a fairness property
The server replays each program and estimates time from joint travel. Client hardware cannot change the result, which is necessary if a score is used competitively or diagnostically.
Method
How the adaptive layer is built
After every attempt the platform re-estimates how the learner is doing and picks the next challenge to match. Finish comfortably and the next one is harder; struggle and it steps back. One level per student, not one level per class.
Model
Two-parameter logistic
Ability θ on a logit scale, item difficulty b, discrimination a. The guessing parameter is fixed at zero: the response space is a program, not a set of choices, so there is nothing to guess into. Estimating a third parameter here would fit noise.
Response
The program is the response
A learner’s Program IR is replayed server-side and scored. The normalized score is remapped around the item’s mastery threshold τ so that “above τ” and “mastered” coincide, then passed to the estimator. The raw score is persisted separately.
Selection
Information, then exposure control
Candidates are ranked by information at the current θ and then exposure-capped, so learners at the same level do not all receive the same items and the bank is not burned through.
Generation
Difficulty as a target, not an outcome
Item families use features that predict difficulty, including clearance, reachability strain, budget pressure, and loop structure. Each candidate is solved for a target b and rejected during generation if the reference solver finds no solution.
Calibration
Provisional until it has evidence
A new item enters provisional: exposure-capped and excluded from ability updates that count. Once it has responses, difficulty is refit by Newton iteration on the marginal likelihood with θ held at posterior means.
This version treats ability as one composite θ and records dimension tags on every response for reporting. A genuinely multidimensional model is a future step, not a claim made here.
Validity
An item nobody can finish measures nothing
Drawing a target haircut first and hoping the arm can reach it does not work. One early challenge asked for 91 pieces of hair when the arm could reach only 20. It was impossible, yet the level gave the learner no warning.
Now the solver runs first. A candidate target is swept against the collision-free joint space, then handed to a reference solver; anything the solver cannot finish is rejected at generation and never served. The solution becomes the level’s reference cost and reference time.
The reachable set is the union of every voxel the tool can contact while sweeping the collision-free joint space. Because that sweep is expensive, it is computed once and cached as a fixture. Tests then verify that the reference solution can be replayed and that no target lies in a dead zone.
The failure that motivated it
A learner could mistake that structural impossibility for personal failure. The ability estimate would then treat the failure as evidence.
Reproducibility
What a result depends on
Deterministic generation
Hairstyle generation is a pure function of the challenge configuration. The same configuration always yields the same ordered voxel set, with no randomness in the pipeline.
Server-side replay
Competitive scores come from replaying the submitted program on the service, never from a number the browser reports.
Estimated, not measured, time
Execution time is derived from joint travel and configured speeds, so it is a property of the program rather than of the machine that ran it.
Frozen wire contract
The app and service share the same scoring and program types. Protocol changes are additive, so older clients keep working.
Elsewhere
Papers and people
Publications
Conference papers, patents, and BibTeX from the group behind HCR, including work on the adaptive learning and testing system used by this platform. The collection is maintained on the Pmine Research site.
The team
Pmine Research at Field of Dream Studio builds HCR. Full profiles are available on the group site.
JHM-SRAS 2026
Jiangsu, Hong Kong, and Macao meet at PolyU for eleven days on intelligent service robotics: four days building low-cost robotic arms with block-based programming, then seven days of papers, exhibition, and lab tours.