If one scout drone feeds ground treaters in a pipeline, can a held fleet beat inline weeding robots?
research/osrobot-pipeline
If one scout drone feeds ground treaters in a pipeline, can a held fleet beat inline weeding robots? One scout keeps only 1.0 to 1.3 treaters busy; overlapping the stages cuts the held-pair penalty to 1.04 at 0.72 ha fields and 0.86 at 1.62 ha, still losing on small fields. Assumed prices; scheduling over osrobot-pairprice's fits.
What it found
A stdlib scheduling follow-up to osrobot-pairprice using its fitted per-field minutes (no new simulation; every price an assumed round number). Set-up and battery swaps make the scout nearly as slow as the treater, so one scout keeps only 0.99 to 1.33 treaters busy and the cheapest pipeline is always one scout with one treater. Overlapping the stages cuts the held-pair penalty to 1.04 times the inline fleet at 0.72 ha fields (32 fields) and 0.86 at 1.62 ha, from 1.34 and 1.16; at 0.18 ha it stays 1.48 to 1.81 times. Extra treaters finish sooner but raise cost per hectare (13.92 to 24.56 USD for 1 to 5). Stylised: deterministic, equal fields, no weather or failures.
A figure from its own output
Paper, code and results
More in Open-source purpose-built robots
- Is a scout drone plus a camera-less ground treater cheaper per hectare than one weeding robot, and from what field size?Is a scout drone plus a camera-less ground treater cheaper per hectare than one weeding robot? With assumed prices and each machine charged for its own minutes, break-even is near 0.5 ha and the pair is about 0.7 times the cost per hectare at 5 ha; held for the whole job it never wins; flips with utilisation, labour and battery life. Stylised, nothing built.
- Does a fast scout platform plus a ground treater beat one weeding robot, and does a mud-aware tour matter?Does a fast scout platform plus a ground treater beat one weeding robot? At matched kill (0.343) a 3 m scout swath finishes in 26.6 min against about 42 min for one inline robot, but a 1 m swath takes 47.3 min; a mud-aware tour changes nothing (stylised, simulation only).
- Does a better tour rescue scout-then-treat weeding?Does a better tour rescue scout-then-treat weeding? 2-opt cuts the tour from 539 m to 468 m but a field still takes 56.1 min against 51.7 min treating inline; scouting loses at every weed density tried (stylised, simulation only).
- For a weeding robot with a cheap camera, is it better to treat weeds as it drives or to scout first and come back, and what does a weak sensor cost?For a weeding robot with a cheap camera, is it better to treat weeds as it drives or to scout first? Scouting then touring killed the same share (0.399 against 0.401 at 1 m/s) and was slower in all 23 paired rows; lane overlap repays a weak sensor (0.40 to 0.86 killed) at 3.5 times the time. Stylised, constants assumed.
- Which open hydraulic circuit gets a small excavator closest to stock energy use for the least money?Research programme (proposal stage): pump-shaft energy of fixed open-centre, load-sensing and ideal independent-metering circuits on an illustrative excavator cycle; over 2,916 swept cases load-sensing uses a median 47% of open-centre energy and metering 80% of load-sensing, an ordering that holds by construction; on 20,000 random cycles in a machine envelope load-sensing is a median 44% of open-centre; no measured cycle, margin or price is sourced yet
- Collision avoidance that obeys the rules of the road: velocity obstacles against rule-based and sampling MPC for a small boatOn 115 scripted COLREGs encounters a rule-aware velocity-obstacle planner passes at least 100 m in 95.7% (rule-based 77.4%, sampling MPC 84.3%), but its safe share falls to 82-85% when the target's velocity estimate is off by 0.3 to 0.6 m/s, and the stand-on vessel's action range matters most: 91% at 200 m, 99% at 600 m
All 44 Open-source purpose-built robots projects →
Other research topics
Proper scoring & elicitation28Effort & peer prediction9Verification games & slashing19Detection & tolerances14Markets 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 robots44The software-engineering loop43Adversarial human imitation41Open-source residential construction41Autonomous multi-angle capture34Sensing, imaging and 3D31Energy supply and dense energy storage39Robot societies among people31Agentic co-design of low-cost robots27Laundry-hanging robots25Machine learning on qualitative data23Robots on the building site27Civilizations & group psychology1History & human sciences1
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.