Presentations of speakers in pdf
- Pieter Abbeel: Lec1, Lec2, Lec 3
- Roman Barták: Lec1, Lec2, Lec 3
- Stan Birchfield: Lec1+Lec2+Lec 3
- Václav Hlaváč: Lec
- Hans-Georg Stork: Lec, Text
- David Vernon: Lec1, Lec2, Lec 3
- Michael Zillich: Lec1, Lec2
Monday July 28, 2014 – Preparation, Day 0
- The project TRADR and the project CloPeMa might organize meetings of their project researchers (not open to public)
- 13:00 (only for member of the TRADR project) Summer school Preparatory meeting (room G205)
- 18:00 – 20:00 ReaRW 2014 Registration, ReaRW 2014 Ice-breaker party in Room G205.
Tuesday July 29, 2014 – ReaRW Day 1
- 8:30 – 8:50 Registration
- 9:00 – 9:25 – ReaRW 2014 Opening, brief intro of each participant
- 09:25 – 09:50 (L01) V. Hlavac: Robotic representation, reasoning and learning – ReaRW 2014 assignment
- 10:00 – 10:50 (L02) P. Abbeel 1: Motion planning and reinforcement learning in robotics
- 11:00 – 11:50 Lecture 3 (L03) R. Barták 1: Robot world representation and reasoning in it
- 11:50 – 13:00 Buffet lunch at the venue (provided), poster session
- 13:00 – 13:50 (L04) P. Abbeel 2: Motion planning and reinforcement learning in robotics
- 14:00 – 14:50 (L05) R. Barták 2: Robot world representation and reasoning in it
- 15:00 – 15:50 (L06) P. Abbeel 3: Motion planning and reinforcement learning in robotics
- 16:00 – 16: 20 Discussion of the day
- 16:20 – Informal program, probably a guided walk in Prague city center.
- 19:00 – Networking dinner at Strahovsky pivovar
Wednesday July 30, 2014 – ReaRW Day 2
- 09:00 – 09:50 (L07) R. Barták 3: Robot world representation and reasoning in it
- 10:00 – 10:50 (L08) D. Vernon 1: Cognitive robotics
- 11:00 – 11:50 (L09) S. Birchfield 1: Perception – action loop in practice
- 11:50 – 13:00 Buffet lunch at the venue (provided), poster session, host institution demos
- 13:00 – 13:50 (L10) H.-G. Stork: Robots, Research, and Responsibility
- 14:00 – 14:50 (L11) D. Vernon 2: Cognitive robotics
- 15:00 – 15:50 (L12) M. Zillich 1: Knowing and acting in an uncertain world – a practical view
- 16:00 – 16: 20 Discussion of the day
- 16:20 – Informal program, probably a small trip in the Prague green finished in a nice pub
Thursday July 31, 2014 – ReaRW Day 3
- 09:00 – 09:50 (L13) D. Vernon 3: Cognitive robotics
- 10:00 – 10:50 (L14) S. Birchfield 2: Perception – action loop in practice
- 11:00 – 11:50 (L15) M. Zillich 2: Knowing and acting in an uncertain world – a practical view
- 11:50 – 13:00 Buffet lunch at the venue (provided), host institutions demos
- 13:00 – 13:50 (L16) S. Birchfield 3: Perception – action loop in practice
- 14:00 – 14:50 (L17) M. Zillich 3: Knowing and acting in an uncertain world – a practical view
- 15:00 – 15:50 (L18) M. Achtelik: MAVs in real applications – State of the art and limits
- 16:00 – Close of ReaRW 2014
Friday August 1, 2014 – Post Summer school activities – Day 4
- 08:30 – 16:00 (only for member of the TRADR project) Discussion, Summary and Evaluation of the Summer School (Room G205)
Talk titles / abstracts
Title: Motion planning and reinforcement learning in robotics
Abstract: This tutorial will cover optimization-based approaches for motion planning, and contrast them with more traditional approaches. This will include a tutorial introduction to sequential convex optimization, and the extension from state space planning to belief space planning. This tutorial will also cover reinforcement learning approaches that are a good fit for robotics, with an emphasis on policy search methods.
Title: Robot world representation and reasoning in it
- Formal models: predicate and first-order logic, Boolean Satisfiability (SAT), Constraint Satisfaction Problem (CSP), temporal models
- World and planning models: situation calculus, set representation, predicate representation, state-variable representation, finite-state automata, planning with time and resources, task networks
- Reasoning techniques: search and inference, DPLL, constraint propagation
- Planning techniques: state-space planning, plan-space planning, translation to SAT/CSP, HTN planning, heuristics, control rules, adding time and resources
- Practical planning: planning systems, PDDL, ANML
- Challenges: real world vs abstract world, open worlds, uncertainty, plan execution
Title: Perception – action loop in practice
- Traditional sense-plan-act vs. behavior-based approach
- Direct mapping from perception to action
- Interactive perception and active sensing
- Objects of interest come from model (in a logical sense), need for segmentation from percepts
- Evolving (logical) model in changing robot world
- Hybrid approach = putting it all together
- Use cases: mobile robot navigation, piece of garment perception and manipulation, map building in mobile robotics
Title: Robotic representation, reasoning and learning – ReaRW 2014 assignment
Abstract (a single 25 min talk):
In this introduction talk, I will formulate the summer school scope and goals.
I will mention the AI and machine learning (pattern recognition) attempts to provide the high-level understanding/control tools to robots. I will also explain why it has been so difficult. I will mention theoretical troubles coming from mathematical logic even for the crisp robot world. It is also rather challenging to induce objects and their semantics from observing real world and acting in it. Dynamical changes and possible interventions of other agents in the robot world make the situation even more complicated.
Finally, I will formulate the expected performance of the high-level modules in projects TRADR (multi-robot-based situation awareness for Urban Search and Rescue scenarios) and CloPeMa (dual-arm manipulation with soft objects as pieces of garments).
Title: Robots, Research, and Responsibility
Abstract (a single 50 min talk):
- This is going to be a meta-research talk of a philosophical nature.
- The speaker retired from the European Commission in early 2012. He worked for many years as a project officer in the area of cognitive systems and robotics. He shaped the calls for project proposals in several EU frameworks (i.e. seven years long planning periods).
- Instead of the the abstract, have a look at his contribution.
- The aim of the talk is to stimulate the discussion among the ReaRW participants.
Title: Cognitive Robotics
These three talks look at the link between artificial cognitive systems and robotics. The first talk will examine the capabilities of a cognitive system – goal-directed action, perception, learning, prospection, and adaptation – before explaining the different approaches that people take to modeling cognition, ranging from symbolic artificial intelligence to connectionism and dynamical systems theory. This foundation sets the scene for the second talk which looks at cognitive architectures, the essential foundation of any cognitive system. We will discuss three different architectures – Soar, Darwin, ISAC, and CLARION – focusing on how they have been applied in robotics. The third talk then addresses in more detail some of the core components of a robotic cognitive architecture, including different forms of memory, development and learning, knowledge representation and reasoning, symbol grounding, prospection and anticipatory action, i.e. planning. We will conclude the third talk by looking briefly at human robot interaction and the role of social cognition in developing robots that can interact effectively with people. While these talks mainly deal with issues at a conceptual level, we will refer to specific computational models where possible to illustrate how some researchers have realized various aspects of cognitive robotics and we will highlight the many challenges that remain.
Title: Knowing and acting in an uncertain world – a practical view
Lecture 1: Robot knowledge
- Task related knowledge: domain knowledge, process knowledge, and dealing with incoming percepts
- Types of representations: symbolic, graphical models, etc.
- Fully symbolic vs. fully reactive/sensor-motor
Lecture 2: Reasoning and acting
- symbolic planning (Roman: How much about symbolic planning will be covered in your lecture?) Actually, my [Roman] idea was to cover automated (symbolic) action planning in “full” but without talking much about plan execution/acting.
- PDDL, STRIPS, extensions to handle time and uncertainty
- graphical models and probabilistic planning
Lecture 3: Use cases, lessons learned, best practices
- A list of real world example solutions from various collaborative projects, their challenges and problems (magic “probabilities”, open worlds, etc.)
- Possibly an exercise with a symbolic planner in ROS
Title: MAVs in real applications – State of the art and limits
MAVs are already used in many applications like surveying and industrial inspections. The talk will give an overview of possible applications as well as technical limitations. The gap between end users’ requirements and todays possible applications links directly to ongoing research in projects like TRADR. In the second part of the talk Ascending Technologies UAV platforms are presented as basis for the TRADR hardware decision.
Responsible: V. Hlavac, email@example.com
Vaclac Hlavac last modified on: 31.07.2014, 16:08