Open-source purpose-built robots
How far can an underwater vehicle dead-reckon before it must surface? DVL drift per kilometre
research/marine-dr
Drift per metre is sqrt(s^2 + (2 sin(b/2))^2) for DVL scale error s and heading bias b (simulation within 1%); with a 0.2% DVL, 1 degree of heading bias drifts 8.7x more than the DVL itself, and a 3 m surface fix every 1 to 2 km holds the 20 km worst-case error to 11 to 15 m against 64 m with no fix
5 tests
What it found
Milestone 6 of track 23 (marine-dr), filed under the marine-hull programme. A stdlib 2-D straight-track dead-reckoning simulator with a DVL scale error (Teledyne Workhorse Navigator bottom-track accuracy 0.2 to 0.4%, from a datasheet excerpt), a heading bias and gyro random walk (assumed, swept) and periodic surface fixes. Check: end error equals the analytic drift rate within 1%. Result: the heading reference dominates unless better than about 0.115 degree; a 10 m error is reached after 3.8 km at 0.2% and 0.1 degree.
Monte Carlo over 200 runs of 20 km: median worst error 64 m with no fix, 15 m with a fix every 2 km, 11 m every 1 km, and the 3 m fix noise floor below that. Straight track only; no currents, DVL dropouts, USBL or surfacing energy; gyro noise and fix quality are assumed, and a first run with an unrealistic gyro noise was discarded. Preliminary; simulation only.
Research log
Ongoing: each run adds a batch and an entry, newest first. Source: LOG.md.
2026-11-02 · Milestone 1: dead-reckoning drift per kilometre and the surfacing interval (hypotheses written before the sweep)
- Inputs: DVL bottom-track velocity accuracy +-0.4% +-0.2 cm/s (WHN 300) and +-0.2% +-0.1 cm/s (WHN 600/1200); single-ping sd 0.3 cm/s at 1 m/s (Teledyne Workhorse Navigator datasheet, via the BODC copy https://www.bodc.ac.uk/data/documents/nodb/pdf/rdi_workhorse_nav_ds_lr.pdf; read as search-result excerpt, PDF not opened here). Treated as a 1-sigma scale error and swept 0.06% to 0.4%. Heading bias 0.05 to 1 degree and gyro random walk are NOT sourced: swept. Surface GNSS fix sd 3 m: assumed.
- H1: simulated end error on a straight track equals L*sqrt(s^2 + (2 sin(b/2))^2) within 1% (tested).
- H2: at 1 deg heading bias the heading term dominates the 0.2% DVL scale error by more than 8x (analytic: 1.745% vs 0.2%); the DVL spec alone is not what sets drift unless the heading reference is better than about 0.11 degree.
- H3: holding a 10 m bound with the WHN 600 scale (0.2%) and 0.1 degree bias needs a surfacing fix every 2.2 km or sooner (analytic: 10/hypot(0.002, 0.001745)).
- Outcome (
experiments/results.txt, PYTHONPATH=src python3 experiments/run.py, seconds). H1 held (tests, within 1%). H2 held: 1 deg bias gives 1.75% vs 0.2% DVL scale, 8.7x; break-even bias is 0.115 degree. H3 was wrong because I mis-did the arithmetic when writing it: hypot(0.002, 0.001745) = 0.00265 gives 3.8 km (table: 3.77 km), not 2.2 km. Logged as my error, not a model finding. - Monte Carlo, 20 km, 200 runs. First run used a gyro random walk of 5e-4 rad/sqrt(s) (about 4 degrees over 20 km), which is not a credible heading reference and swamped everything (no-fix median max error 465 m); discarded and rerun at 5e-5 rad/sqrt(s) (about 0.17 deg/sqrt(h); unsourced, tactical-grade order). Rerun: no fix median max error 64 m (95th 178 m); fixing every 2 km 15 m (28 m); every 1 km 11 m (18 m); every 0.5 km 9.6 m, which is the 3 m fix noise floor (max over many fixes), so fixing more often than about 1 km buys nothing at this fix quality.
- Limits: straight track only (no turns, so heading bias is a pure cross-track chord); constant bias; no current-induced water-track error, no DVL dropout, no USBL; the 3 m fix and gyro noise are assumed; DVL spec read from a search excerpt, datasheet not opened. Surfacing energy and time cost not modelled, so "how often to surface" is the error side only.
- Next: add turns (lawnmower), a USBL range aid, and the energy cost of a surfacing from marine-hull's glider/AUV energy model.
Paper, code and results
More in Open-source purpose-built robots
- One big survey boat or several small ones at equal cost? Fleet sizing for a lawnmower surveyOn a line-spacing survey, a fleet of smaller boats bought with the money for one large boat finishes in a median 0.29 to 0.35 of the time and uses a median 0.20 to 0.35 of the energy in the 2,178 to 2,186 of 2,187 cases where it is affordable and wins (never slower)…
- Which body carries a survey payload furthest per kWh? Hull choice for small autonomous boatsResearch programme (proposal stage): energy per survey kilometre of hull types for 100 to 500 kg autonomous boats; with a Michell wave term a catamaran out-ranges a monohull in 3.3% of 6,561 cases, none at equal length; with Slocum coefficients from the gliderflight package a glider never out-ranges a same-mass propeller AUV at equal hotel load (0 of 2,430 cases)…
- Holding a survey line in current and waves: line-of-sight against integral line-of-sight for a small boatPlain line-of-sight guidance leaves a steady cross-track offset of delta*tan(asin(c/U)) in a cross current c (simulation matches it to 1%: 3.77 m at 0.3 m/s, 12 m look-ahead)…
- How much does tacking cost a wing-sail drone, and does a tack-aware router win it back?In 54 simulated 38 km beats of a 4 m wing-sail drone (parametric wind and tide), a router blind to the tack stall loses a mean 4.5% in steady wind (max 9.8% at a 120 s tack), planning with the cost cuts that to 1.0%, and a greedy layline controller loses up to 17.6% with a 0.5 m/s tide and 3 to 25% when the wind shifts
- If a late field loses yield instead of failing a hard deadline, does a scout-and-treater fleet still lose to inline weeding robots?Replaces the hard deadline of osrobot-deadline with a 2% yield loss per leaf stage of delay (Knezevic et al. 2003): the best pipeline costs 1.06 to 1.20 times the best inline fleet at 0.72 ha and 1.49 to 1.51 at 0.18 ha, and at 1.62 ha inline is cheaper when the window closes by 8 h and the pipeline from 12 h
- What does a 2-joint feeding arm cost against a 6-joint one once the whole bill of materials is priced?What does a 2-joint feeding arm cost against a 6-joint one once the whole bill of materials is priced? About 1.8 times, not 3: $66 against $122 at the SO-ARM101 table's $13.89 servo, rising to 2.4 times at a $49 retail servo.
All 50 Open-source purpose-built robots projects →
Other research topics
Proper scoring & elicitation28Effort & peer prediction9Verification games & slashing19Detection & tolerances15Markets as routers13Prediction-market design9Paying contributors8Decentralised training systems23Digital twins & sim-to-real164LLM chip design loops40Robot manufacturing loops53Ambient AI assistant46Self-replicating manufacturing34Simulation-first design64Engineered clothing45Task-specific AI accelerators40Algorithmic chip-design loops42Open-source purpose-built robots50The software-engineering loop43Adversarial human imitation41Open-source residential construction41Autonomous multi-angle capture35Sensing, imaging and 3D33Energy supply and dense energy storage41Robot societies among people32Agentic co-design of low-cost robots28Laundry-hanging robots26Machine learning on qualitative data23Robots on the building site28Civilizations & group psychology3History & human sciences4
Produced with AI assistance (Claude). Results are from the code in this repository and are simulated or stylised unless the abstract says otherwise; nothing here implies affiliation with or endorsement by any lab, researcher or company cited. MIT licensed.