Workshop 1
Harsh environment robotics – Prof. Fumi Matsuno

1. Transfer of Skill from Human to Robot – Kouhei Ohnishi¹
   There are many atypical tasks which are easily accomplished by the human but are very difficult for the robot. In most of such cases, the human skills are indispensable but they can not be digitized nor transferred by the communication line. That prevents the robot to use the human skills in such tasks. The paper shows how we can transfer the human skills to the robot. In the proposed strategy, two types of AI are necessary which are corresponding to human cerebrum and cerebellum. The experimental examples based upon this strategy shows that the robot can accomplish sensitive tasks which are carried out only by the human.
2. Autonomous, Intelligent, and Swarm Control Research Unit – Masayoshi Tomizuka²
3. Fuel cell system development for harsh environment at F-REI – Akihiro IIYAMA, Go Matsuo, Tetsuya Kamihara, Keiji Okada, Atsuko Fukaya, Satoru Imazu, and Masanari Yanagisawa³
   Drones flying in harsh environments are expected to require long flight times and large payloads, which are difficult to achieve with conventional batteries. We are conducting research and development on the application of hydrogen-fueled polymer electrolyte fuel cells to drones, focusing mainly on weight reduction and durability. An overview of the research and development will be reported.
4. Fluid powered robots for harsh environments – Koichi Suzumori²
   Fluid-powered robots (pneumatic or hydraulic robots) that the author has developed to date are introduced. Pneumatics realizes lightweight and compliance, while hydraulics realizes high power density and environmental resistance. They can be used in harsh environments (rain, dust, shock, unknown environment, etc.).
5. Aerial Interception of Multiple Drones by Multiple Distributed Chasers with Batch Deployments – Nontaphat Charuvajana¹, Panithan Rithburi², Sutthiphong Srigrarom³, Boo Cheong Khoo³ and Florian Holzapfel⁴⁴
   In this paper, we investigate the problem of air-to-air drone interception for counter-UAS operations, using multi-agent reinforcement learning, strategic surrounding approaches and sequences of deployments. The intruding drones are modelled by Boid Algorithm, analogous to flocks of birds. The fleet of interceptors (agents) are assigned to intercept or chase the intruders mid-air at designated area in batch deployments. For intercepting or catching purpose, we applied the StringNet strategy and greedy strategy to break, herd and encircle several subgroups of targets. We also apply task allocation algorithm to assign each of the agent to track and look for specific target within the subgroup. This allows better chasing effectiveness, when there are limited numbers of intercepting drones, assume equalled to or smaller than the intruding drones. The heuristic task allocation framework is modeled as matching and optimization problem. The preliminary mix-integer, non-linear problem (MINLP) formulations are based on probability of interception and resource readiness. The interceptors are deployed in sequences of batches allowing follow-up action. The single deployment performs better in term of action time, however, the batches deployments give higher success rate due to clearer task assignments. Preliminary works have shown that the combination of the proposed Hunting, heuristic task allocation and batch deployment performed well for as many as 15 intruders, and by 15 interceptors with 100% interceptions.

---

1. Keio University and F-REI ↩︎
2. University of California Berkeley and F-REI ↩︎
3. F-REI ↩︎
4. [1] National University of Singapore, Computer Science Dept, Singapore
   [2] Technical University of Munich Asia, Singapore
   [3] National University of Singapore, Mechanical Engineering Dept, Singapore
   [4] Technical University of Munich, Institute of Flight Systems Dynamics, Germany ↩︎

Workshop 2