Ideas to pursue

As usual, I’m full of a lot of different ideas I’d like to investigate more deeply and develop into useful things. Among these are:

• Manufacturing robot: I want some kind of precise positioning system that I can use to deposit or remove material. The shortest path to this is likely inkjet printer carriages, but I need to figure out how to physically hook electronics up to their linear encoders, and then I probably need to rig up some kind of PID control to drive them with.
• The bed platform: I need to build a thing in my bedroom to turn it into more of a bedroom and be able to move in. This involves buying or salvaging more steel studs, hooking them together, and putting some shit on top, and then at least some kind of cloth around the edges.
• Extending the calculator: I have a somewhat awesome interactive calculator, but it still needs a lot of work. It would benefit a lot from localStorage, decimal points, and fixing editing so that I can add new numbers next to existing ones, and also from a multitouch UI.
• Multitouch UI: I have a promising prototype that isn’t yet good enough to do anything useful.
• Ice vest: something to enable me to comfortably weather the summer in Buenos Aires.
• Dependency-driven recalculation: crossing make with Apache Spark.
• Graph notation: a general way to represent digraphs as relatively short sequences of symbols.
• Magic Kazoo: a synthesizer you can hold in your mouth and play like a kazoo
• Constraint-based quantitative modeling: a system for building quantitative models that can include things like three-dimensional shapes.
• Ghettobotics: bootstrapping a bitchin electronics lab from garbage.
• Quasicard: a generative form of hypertext that supports exploratory data analysis
• Stuff with optimization: a whole list of things that I should be able to attack with mathematical optimization, including:
  • Dithering.
  • Image approximation under other constraints, such as mosaic tiles or gradient swatches, using a psychovisual model.
  • Or, for example, with an XY robot, an ink model, and a psychovisual model.
  • Melody transcription.
  • Magic Sinewave synthesis (with a given reactive filter and switching losses, say).
  • Sparse approximation of convolution kernels (generalizations of the Hogenauer filter).
  • Kinematic control.
  • Topological optimization for panel cutting.
  • FEM in general.
  • Text layout, as TeX’s badness-minimization approach. But e.g. for JSON.
  • Toolpath optimization?
  • Analog filter design.
  • Relatedly, IIR filter design.
  • Structure from motion, photogrammetry, and structure from shading.
  • Tool choice from mesh model and tolerances — using more precise tools where tighter tolerances are needed.
  • Circuit design from available components (measured e.g. with an M328)
  • Execution planning in e.g. a query planner, where there are multiple possible ways to compute the same result.
  • Index design for a database, which is the same thing taken up a level of abstraction — which indices would minimize query cost?
  • Weld placement on weldments to optimize their strength
  • Automated bricolage from gluing together found objects
  • Mesh complexity reduction with bounded error
  • Curve approximation with minimal numbers of control points
  • List decoding of noisy signals
  • Signal approximation (of e.g. audio signals) from given primitive signals and combining operators and a closeness metric (e.g. a psychoacoustic model). One aspect of this is audiomontage, where you combine prerecorded sounds to produce a desired sound, as in notjustmoreidlechatter.
  • Speech recognition, given a speech synthesis program and a psychoacoustic model.
  • Generating knitting patterns from triangle-mesh 3-D models
  • Generating laser-cutting patterns from mesh models
  • Cost optimization of laser cutting under various constraints, e.g. bounded error to a mesh model, or visual similarity using a psychovisual model.
  • Cache allocation optimization
  • Planning a travel path through a city to minimize cost, discomfort, worst-case or average-case travel time, or some combination
  • Image coregistration
  • Tensegrity design given constraints (e.g. maximum rigidity for comfort, minimum rigidity, minimum strength, impact energy, strength under given loads, support points)
  • Graded-index optical systems with no surface scattering
  • Materials property estimation from experimental results
  • Circuit or electronic component estimation from experimental results
  • Design of linkages or composite materials to produce a given force-displacement curve or force-displacement-time curve
  • Inverse filtering to compensate precisely for known output transducer imperfections, including nonlinearities
  • Also, of course, for input transducer imperfections
  • Optimizing a predictor of the next sample used for compression (including lossless compression) for a given image or sound.
  • Estimating a convolution kernel from two simultaneous signals, such as a stereo recording of someone talking in a church. Basically this is just inverse convolution.
  • Beamforming
  • Toolpath planning not just from a 3-D model but also, for example, to approximate a given grayscale image by, for example, carving plaster.
  • Attributing motivations to human choices, e.g. social network analysis from Facebook “like” data
  • IQlight design from mesh
  • Robot path planning
  • Robot attitude and position information from sensor data — actually SLAM as a whole
  • Most-probable-fault diagnosis from a fault tree and observations
  • Inferring a fault tree from observations
  • Experiment design for fault diagnosis given a fault tree and partial information
  • Learning a model of a controllable system (e.g. a robot arm) in order to control it and know where the controllable zones are
  • Designing a procedure for a controllable system to optimize the important tolerances instead of the less important ones; this broadly includes things like planning to cast surfaces that don’t need precision followed by grinding surfaces that do, but also planning to position workpieces such that the things that need precision are in the zones where your actuator and feedback are most precise
  • Constraint satisfaction for 3-D modeling
• qyap
• a small, safe IRC client
• a daisy-chaining bus for simple programmable electronics