Does searching an arm’s joint count, link lengths and mount, with its controller tuned in the loop, beat a fixed three-link body at the same cost?

What it found

A stdlib model searches link lengths, joint count and mount position of a planar arm, with its inverse-kinematics controller tuned in the loop, against a fixed three-link body at an equal cost budget (300 evaluations, 10 seeds, held-out targets). On a shelf task that needs the tool pointing down, the search reaches 0.95 success against 0.26 for the fixed body with a tuned controller and 0.50 with the mount also tuned; on a long position-only reach the fixed body already succeeds and the search adds nothing. Random search beat a (1+4) evolution strategy here, weakly. Stylised: kinematics and static torque only (no MuJoCo dynamics), and every cost, mass and limit is assumed.

A figure from its own output

Paper, code and results

• README and method
• Experiments and results
• Code
• Tests

More in Open-source purpose-built robots

• How heavy must a purpose-built UAV for targeted observation be, and when does a hover-capable airframe stop being worth its weight?How heavy must a purpose-built UAV for targeted observation be, and when is a hover-capable airframe worth its weight? At 3 km out and 20 min on station a fixed wing closes at 1.27 kg against 2.14 kg for a quadrotor and 2.08 kg for a VTOL wing, and the quadrotor stops closing at 45 min (stylised, assumed constants, no flight stack run).
• Does a steady wind change which purpose-built observation UAV is lightest, and what is wind worth to a multirotor?Does a steady wind change which purpose-built observation UAV is lightest? At 3 km and 20 min on station the fixed wing stays lightest from 0 to 18 m/s (1.27 to 1.50 kg) while the quadrotor goes 2.14 to 7.0 kg; the multirotor's still-air win at 1 min on station ends near 10 m/s. Stylised, steady known wind only.
• How much heavier must an observation UAV be if the wind is uncertain, and is designing to a wind quantile cheaper than a bigger battery reserve?How much heavier must an observation UAV be if the wind is uncertain? Designing to the 95th-percentile of wind ~ N(8, 3) m/s costs the fixed wing +3.7% and the multirotor +11.7% mass, and a bigger battery reserve buys about the same reliability for about the same mass; stylised, one steady wind per mission.
• If an observation UAV only launches when the forecast wind is below a limit, how much does forecast error bring failures back, and what does it cost in scrubbed missions?If an observation UAV launches only when the forecast wind is under a limit, how much does forecast error bring failures back? A perfect forecast turns every failure into a scrub (fixed wing, wind ~ N(5, 4) m/s: 4.6% scrubbed at the 95th-percentile limit); a 2 m/s error gives 6.8% failures at the median limit but 1.1% at the 95th. Stylised, one steady wind per mission.
• If a scrubbed observation-UAV launch can be retried later the same day, how much does persistent wind limit what retrying buys, and does retrying bring failures back?If a scrubbed observation-UAV launch can be retried later the same day, how much does persistent wind limit the gain, and do retries bring failures back? Fixed wing, median-wind limit, perfect forecast: eight slots complete 99.7% of days with independent wind but 71.5% with persistence 0.95; with a 2 m/s forecast error eight slots raise failures from 6.8% to 13.6% of days (23.7% at persistence 0.95). Stylised.
• What is a UAV launch wind limit worth in money, and should the launch threshold sit below the design limit when the forecast is wrong?What is a UAV launch wind limit worth in money, and should the launch threshold sit below the design limit when the forecast is wrong? With an assumed expensive airframe and a 2 m/s forecast error the best threshold sits 1 to 6 m/s below the design limit, worth up to $284 per attempt, with failures at or below 0.1%; with a cheap airframe the best design covers nearly every wind. Stylised, assumed prices.

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.