Does rotor cant still pay once a co-designed quadrotor must hold position?
research/codesign-aero-cant-poshold
Once the vehicle must hold position, does a canted-rotor quadrotor still win? In simulation cant still buys yaw (−6° RMS, 5 of 5 seeds) but not position, and the sideways force a tilted rotor makes under roll/pitch torque breaks designs tuned without it (1.1 m against 0.25 m position error).
What it found
Five earlier codesign-aero-cant* projects deferred position hold. Here the quadrotor gets translation, an outer position loop and the body-lateral force of a canted rotor, F = τ tan α / r per unit commanded roll or pitch torque (closed form checked against an explicit four-rotor allocation). Joint CMA-ES over arm length, propeller size, cant and nine gains (3200 evaluations, 5 seeds, 64 held-out episodes). Cant still cuts yaw error by 5.8° (bound 35°) and 6.0° (bound 80°), better in 5 of 5 paired seeds, while position error is unchanged within noise (0.25 m against 0.22 m with no cant). Designs tuned with the lateral force switched off and then scored with it on reach 1.1–1.2 m position error, worse in 5 of 5 seeds; the force acts as a disturbance (switching it off improves force-aware designs), with an unhelpful sign on the roll axis for this layout. Limits: 2 s episodes, objective weights (0.1 m against 10°) chosen, several seeds stop in local optima, one arm layout, perfect altitude hold; cost model, noise, offset and plant ranges are assumed or inherited, not measured; no hardware.
A figure from its own output
Paper, code and results
More in Agentic co-design of low-cost robots
- What reaction-torque-to-thrust ratio do real small propellers have, and does the best rotor cant angle survive it?What is the reaction-torque-to-thrust ratio of real small propellers, and does the best cant angle survive it? Measured static data for 77 propellers of 3.15–9.85 in give c = CP/(2π·CT) = 0.065–0.13 (median 0.081); in simulation cant still beats no cant in 5 of 5 seeds at every c, by 2.9–4.7°, and the joint design picks 36–41°.
- Is rotor cant just a cheap way to buy yaw authority that bigger motors would buy as well?Does rotor cant still help once motor headroom is a design variable? With thrust-to-weight free in 1.5–4 (no cant picks the 4.0 ceiling), cant to 35° still beats no cant in 5 of 5 seeds at each of three measured propeller c values, by 1.6–3.4°; headroom alone recovers 2.5–3.5° of the 3.8–4.7° the parent credited to cant.
- Was the thrust-to-weight ceiling of 4 what made rotor cant look good?Does raising the TWR ceiling from 4 to 6 and 8 erase the cant advantage? No: cant to 35° still beats no cant in 5 of 5 seeds at three measured propeller c values, by 1.5–3.0° at ceiling 8 (1.6–3.4° at 4); the optimiser stops near TWR 5 without cant.
- Does a neural residual help a quadrotor attitude controller once roll, pitch and yaw are coupled?Does a residual network or analytic gyroscopic compensation beat per-axis PID on a coupled three-axis quadrotor? No: the residual is worse than PID in 26 of 30 pairs (+0.32° held out) and compensation changes the error by 0.001°; yaw authority sets the error.
- Does a small neural roll controller beat the three-gain PID on a randomised quadrotor, and does it change which airframe is best?Does a 31-parameter neural roll controller beat the three-gain PID on a randomised quadrotor? Not at equal evaluations (7.7° PID against 9.0°); with 10× the evaluations it only matches it (7.6°), and the airframe ranking is unchanged (Spearman 1.00).
- Does a neural residual on top of the tuned PID get a quadrotor roll controller below the PID error floor?Does a zero-initialised neural residual on the tuned PID beat the PID on a randomised quadrotor? No at equal evaluations (7.68° against 7.58°, better in 7 of 18 pairs); the airframe ranking is unchanged (Spearman 1.00).
All 27 Agentic co-design of low-cost 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.