[ICON Fall25 Seminar] Mark Yim (UPenn) #robotics #autonomy #grasp

In this talk

1. 5:22

   Modular robot designs and tradeoffs

   Yim describes chain, lattice, and hybrid robots built from repeatable modules. Chain designs work well as articulated arms, while lattice designs change shape more readily; hybrid systems combine those strengths. Their promised versatility comes with practical costs: modules can fail, raise system expenses, and make a robot less optimized for any one task.

2. 12:32

   Limits of miniaturization and strength

   Shrinking modules can increase the number of possible configurations, but small processors and sensors constrain capability, and total system cost matters more than the price of one module. Yim contrasts a simple PolyBot module used in hundreds of demonstrations with a more elaborate $3,000 version used in only a few. He also explains how weight limits extended robot arms and describes scissor mechanisms and distributed motors as possible ways to scale strength.

3. 33:07

   Docking and choosing robot shapes

   Reconfigurable robots need reliable ways to connect modules despite alignment errors. Yim compares docking designs including the DARPA TESS hook-and-loop system, ATRON latches, and electropermanent magnets, which can switch magnetic force with a brief pulse. He also outlines a task-first approach to robot design: assess the environment and required motion, then generate and test a suitable configuration.

4. 46:27

   Responsible innovation in robotics

   Yim frames responsible innovation around anticipating effects, examining assumptions, including affected groups, and changing direction when needed. He asks researchers to imagine their work becoming widely successful, then consider potential harms and ways to reduce them. Examples include a cancer sensor being restricted by a company, a sign-language robot unwanted by the community it aimed to help, and risks to jobs, security, and privacy.

5. 1:05:17

   Incentives, consent, and public impact

   The discussion raises how academic citation pressure and viral attention can reward impressive demonstrations over useful work or careful scrutiny. Participants also question consent when self-driving cars collect data from pedestrians and when artists' work is used to train generative systems. Yim says researchers and institutions need to recognize harms and be ready to respond, while acknowledging that better ways to make that happen remain unclear.