Machine learning for tactile manipulation Jan Peters with Filipe Veiga, Herke van Hoof, Oliver Kroemer, Roberto Calandra, Tucker Hermans, Yevgen Chetobar, Yilei Zheng, Zhengkun Yi Intelligent Autonomous Systems Dept. of Computer Science Technische Universität Darmstadt Interdepartmental Robot Learning Group Depts. of Autonomous Motion and Empirical Inference Max Planck Institute for Intelligent Systems All work in this talk is part of the EU FP7 ICT Project “Tactile Manipulation” (TACMAN). Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS|
Why don’t we have personal robots yet? Manipulation appears so easy: •all open loop •all hard coded •recently: just add a little vision Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 2
So what was hard-coded and pretended in this TV commercial of the first robot ever? Predict material properties Predict slip and control it Efficient teaching of Predict slip and use it Efficient tactile robots by imitation & RL for gripping exploration of surfaces We need learn to act using tactile sensing! Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 3
Property recognition How can we learn to act using tactile sensing? Tactile exploration 1. What can we learn to recognize from tactile interaction? 2. How can we efficiently explore through touch? Predict & Control Slip 3. How can we learn to control slip from touch? 4. Can we obtain modular grip control from single finger slip control? Grip Control by Slip Control 5. How can we self-improve manipulation? Efficient teaching Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 4
Property recognition How can we learn to act using tactile sensing? Tactile exploration 1. What can we learn to recognize from tactile interaction? 2. How can we efficiently explore through touch? Predict & Control Slip 3. How can we learn to control slip from touch? 4. Can we obtain modular grip control from single finger slip control? Grip Control by Slip Control 5. How can we self-improve manipulation? Efficient teaching Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 5
What can we learn to recognize from tactile interaction? •Object and material recognition is crucial for manipulation •Allows recognition at the point of contact and type of possible interaction Predict material properties •Allows for the absence of accurate models and vision ➡ How much on material properties can we predict from tactile sensing? Hoelscher, J.; Peters, J.; Hermans, T. (2015). Evaluation of Janine Tucker Interactive Object Recognition with Tactile Sensing, Proceedings Hölscher Hermans of the International Conference on Humanoid Robots. Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 6
Experiment – Object Recognition Collect data for every object ¡ Extract features ¡ Select training and validation data ¡ Train a classifier Use features to predict classification Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 7
49 Objects Plastic Metal Paper Fabric Ceramic Stone Wood Sponge Miscellaneous Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 8
Confusion Matrix Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 9
Confusion Matrix Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 10
Materials classification Material classification, movement concatenation, linear SVM Accuracy: 97.55% for 49 objects with textbook methods Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 11
Property recognition How can we learn to act using tactile sensing? Tactile exploration 1. What can we learn to recognize from tactile interaction? 2. How can we efficiently explore through touch? Predict & Control Slip 3. How can we learn to control slip from touch? 4. Can we obtain modular grip control from single finger slip control? Grip Control by Slip Control 5. How can we self-improve manipulation? Efficient teaching Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 12
How can we efficiently explore through touch? • Accurate object shape knowledge provides import information for complex tasks such as grasping. • Vision-based methods suffer from limitations such as the available illumination and are not applicable when the object is not visible or occluded. Efficient tactile exploration of surfaces • When modeling an object using tactile sensors, touching the object surface at a fixed grid of points can be sample inefficient. Zhengkun ➡ Efficiently exploring such object knowledge is key Yi for many tasks! Yi, Z.; Calandra, R.; Veiga, F.; van Hoof, H.; Hermans, T.; Zhang, Y.; Peters, J. (2016). Active Tactile Object Exploration with Gaussian Processes, Proceedings of the IEEE/RSJ Conference on Intelligent Robots and Systems (IROS). Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 13
Efficient Active Tactile Object Exploration with Gaussian Processes • Use Gaussian Processes to model object surfaces. • Choose as covariance function the squared exponential. • Inspired by Bayesian optimization (BO). • The acquisition function is defined as the predicted standard deviation. • Use DIRECT to find the approximately optimal solution. (Jones et al. 1993) Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 14
Experiment: Using a Real Robot BioTac Experimental setup • The object is fixed to a vertical surface. • The rectangular zones are to be reconstructed. Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 15
Experiment: Fast convergence… Correlation coefficient True function Reconstructed function The correlation coefficient almost converges to 1 much faster when using the active touch approach. Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 16
Property recognition How can we learn to act using tactile sensing? Tactile exploration 1. What can we learn to recognize from tactile interaction? 2. How can we efficiently explore through touch? Predict & Control Slip 3. How can we learn to control slip from touch? 4. Can we obtain modular grip control from single finger slip control? Grip Control by Slip Control 5. How can we self-improve manipulation? Efficient teaching Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 17
How can we learn to control slip from touch? • Tactile sensing gives direct insight on the state of the object and does not suffer occlusions. • Discrete events greatly influence the manipulation tasks. • Slip between fingertip and object often results in Predict slip and control it task failure. ➡ Can we predict the onset of slip and prevent/ control it before it happened? Veiga, F.F.; van Hoof, H.; Peters, J.; Hermans, T. (2015). Stabilizing Novel Objects by Learning to Predict Tactile Slip, Proceedings of the IEEE/RSJ Filipe Veiga Conference on Intelligent Robots and Systems Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 18
Slip prediction as classification problem c t + τ f = f ( φ ( x 1: t )) Slip prediction c t + τ f ∈ { c slip , c non _ slip } •as classification •with prediction horizon τ f τ f = 0 •Just slip detection Features for classification • Single element feature φ x ( ) = x t 1: t • Delta feature with ( ) = x t , Δ x t [ ] φ x Δ x t = x t − x t − 1 1: t • Time window feature φ x ( ) = x t −τ : t 1: t Classifiers • Linear SVM, Random Forests Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 19
Experiments in slip prediction Slip prediction Data set F score Data collection τ f Prediction horizon No significant drop in performance! Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 20
Grip stabilization by slip prediction c t + τ f = f ( φ ( x 1: t )) Features Slip Grip based on Prediction as Stabilization Tactile Binary Control Based information Classification on Slip Signal φ x ( ) " F N [ t ] + σ ˆ F N [ t ] if slip $ 1: t F N [ t + 1] = # F N [ t ] otherwise $ % Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS|
Grip Stabilization Experiment • Objects pinched against a vertical table. • Robot attempts to move away from the with random velocity. • Grip Stabilization controller triggers when slip occurs. • Controller stays active until object is successfully stabilized or maximum time duration is reached. Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS|
Grip Stabilization Experiment Success rate Prediction horizon Slip prediction horizon greatly increases stabilization success rate. Jan Peters | Intelligent Autonomous Systems @ TU Darmstadt|Robot Learning @ MPI-IS| 23
Recommend
More recommend
