Vincenzo Ceccarelli GrimaldiCluster · Physical AI
Menu

Physical AI & Robotics · research and evidence

Research program

A research claim exists here only with baseline, dataset and metric named. Until an experiment has run, the state is OPEN and no result is quoted. No performance claim is made without benchmark evidence.

TopicRelevanceBaselineDatasetMetricState
Vision-language-action modelsScene understanding for mixed-SKU picking and inspection.Classical detector + hand-written grasp rulesPalletizing fixture set (to be published)Successful picks, latencyOPEN
World modelsPredict pallet state and occlusions before a place.Deterministic pallet-state tracker in palletizerSimulation tracesPlacement error, stability prediction accuracyOPEN
Diffusion policiesGrasp and place trajectories.Sampled motion plannerSimulation tracesCycle time, collision rateOPEN
Imitation learningOperator demonstrations for irregular cases.Rule-based pattern managerRecorded demonstrations (none yet)Intervention rateOPEN
Reinforcement learningSequencing under changing SKU mix.Heuristic sequencerSimulated order streamsCycles/hour, changeover timeOPEN
Model predictive controlConveyor–arm coordination.Fixed-rate orchestrator loopSimulationRobot utilisationOPEN
Task and motion planningJoint choice of placement and trajectory.Sequential task-then-motionFixture setPlan success rate, planning timeOPEN
Sensor fusionThe perception layer itself.Single-sensor detectionKITTI-calibrated sequences + own benchmarkPrecision, recall, localisation errorOPEN
Uncertainty estimationKnow when not to pick.Confidence thresholdBenchmark with degraded subsetsFailure rate at fixed recallOPEN
Sim-to-realTrust the simulator’s KPIs.Simulation onlyPaired sim/real runs (needs a cell)KPI gap sim vs realOPEN
Active learningWhich failed picks to label.Random samplingTelemetryLabel cost per point of recallOPEN
Failure recoveryRECOVER step of the mission.Stop and call operatorFault logInterventions avoidedOPEN
Safe explorationLearning that cannot leave the constraint set.No learning in the loopSimulationConstraint violations (must be zero)OPEN

Every reported experiment will carry: baseline, dataset, metric, confidence interval, hardware, latency, failure cases, generalisation, ablation, and a reproduction path from a clean clone.

External intelligence

For every important development the log answers five questions, then gives a verdict.

  1. What is new?
  2. Is it actually better?
  3. Can we reproduce it?
  4. Does it create commercial advantage?
  5. Do we buy it, build it, partner, or ignore it?

Log empty as of the first weekly cycle. An empty log is stated, not hidden; a list of venues is not intelligence.

Watchlist

Customer validation

Conversations recorded: 0. Validation is money, a signed pilot or a committed design partnership — never a compliment. Segments and the seven questions are fixed; answers are logged here as they arrive.

Segments

  • System integrators
  • Manufacturers
  • Warehouses
  • Packaging companies
  • Logistics operators
  • Inspection companies
  • Utilities

Questions

  1. What task costs you the most labour?
  2. How often does it occur?
  3. What is the current intervention rate?
  4. What does downtime cost?
  5. What is the current automation solution?
  6. What prevents further automation?
  7. What would a successful pilot be worth?