Bug fixes and minor updates - 2023 edition

Bugs

• Fixed logging in Pareto evaluation methods, now correct reference to simulation instance variable
• eval_parameter helper function now correctly supported predict = false case
• If simulation doesn't start at 0, then simulation time in agent's history is now correctly prefilled

Improvements

• Agents' doNegotiation now predicts state trajectory xPred as well as eval predictions and stores them in the status

Bug fixes and minor updates

Bugs

• Fixed a typo in lexicographic extreme point method
• Comparison of equality between extreme point solutions now with tolerance instead of exact comparison
• Fixed ParetoController breaking, when a cost function with weight = 0 was added to the controller
• Fixed prediction horizon status.horizonTime not being set in negotiation
• Fixed fixed eval predictions accidentally being cleared in previousStatus when eval function is scenario dependent

Improvements

• The simulation results directory can now be set freely, by passing it as a an argument when creating a Simulation
• Slight cleanup of Pareto code
• Front determination methods will now print progress to log
• Added section on how to cite PARODIS to the readme

Virtual history, sensible constraints and a fancy crane

Bugs

• Constraint on du(0|k) = u(0|k) - u(k-1) is now considered in delta constraints, u(k-1) is automatically provided
• Pareto optimization now works if no additional slack variables are defined
• Figures which shall not be live plotted will no longer be opened when simulation starts with livePlots = true
• Fixed FPBI front determination scheme storing broken fronts
• Fixed wrong handling of extreme points in AWDS and ASBI front determination schemes
• Ugly error when no point was selected for plotting trajectories in Pareto interactivity tool
• sdpvars created by createModel and the controllers are now always full n x m matrices, even if n == m

Improvements

• Introduced the virtual history, which tracks the predicted system behaviour in each timestep, while the history tracks the actual system behaviour, caused through state measurement or real vs. predicted disturbances
• PARODIS now delivers a pre-built cost function for the classical min xQx + uRu control case in the functions directory called LQRCostFunction
• Added another example, a container crane system, which can be found in examples/crane
• Added a hierarchical MPC example, namely the one presented in the CCTA paper on the PARODIS framework
• Refactoring, mainly regarding how the history (and virtual history) is updated, better code encapsulation
• Pareto front determination schemes now log errors to agent log
• Added warmstarting option for (nonlinear) solvers in explicit and pareto controllers, can speed up solver time for some NLPs

Changes

• Arguments in createModel and the controller constructors for the number of scenarios to be considered are now optional and default to 1
• Argument scenarioDependent in controller.addParam is now optional, defaults to true
• Simulation option doCopyCallingScript is now false per default

Again, lots of fixes and some Pareto improvements

Bugs

• Fixed broken delta constraints for du
• Bug fixed in lexicographic extreme point method that would result in the same extreme point twice
• Fixed the ParetoController accidentally storing data in the status struct wrapped in cell arrays

Improvements

• Added new Pareto front determination scheme, Adaptive Search Boundary Intersection (ASBI)
• getSlacks and getConstraints methods in CostFunction are no longer abstract and have default implementations
• Controller now checks if getSlacks correctly returns a struct
• RoC metric now ignores concave parts on fronts
• Refactoring

Lots of fixes

After about two months of fixes culminating on the develop branch, we're ready for another release!

• Fixed critical bug where non-implicit constraints caused crash when using ExplicitController/ParetoController
• Lots of refactoring in ParetoController
• Added checking for invalid initial state x0
• Better warnings when things go bad in multi-agent use
• Fixed bug where sdp variables where passed to a cost function's evaluateHorizon method instead of numerical values

Initial Release

Initial public release of PARODIS 🎉