Kashyap lays out a range of arguments for why open source is the way to go, using analogies from software industry: Linux vs. Windows NT.
A good time to check the essay of Bill Gurley p3institute.substack.com/p/from-open-... #ssad2026

Can we have an open-source wave in automotive like the one on LLMs and mimic the fast progress in that area?
There are already quite a few players, but there is certainly room for more. #ssad2026
PS: super proud to see @valeoai.bsky.social on that list

The reality check of current state of AD deployment 😬 #ssad2026

Autonomous Driving then and now:
- 1990s Mercedes had a prototype running at 180kph on German highways
- 2020s a gazillion of prototypes and products out there #ssad2026

Fatma Güney up next speaking on Learning compact state representations for reliable autonomous driving #ssad2026

We start the 2nd #ssad2026 day with @kashyap7x.bsky.social talking about: Making progress measurable - Evaluation and benchmarking of driving policies

Honest open problems in RL for AD and tentative workarounds by
Maximilian Igl #ssad2026

Main approaches for finetuning e2e driving models in close(ish)-loop nowadays #ssad2026

One of the main weaknesses for each RL paradigm: the ego-policy adapts to whoever drives the other cars #ssad2026

Reward design in an RL for AD in a nutshell.
Nice summary by Maximilian Igl #ssad2026

The many trade-offs of setting the reward in RL for AD #ssad2026

A particularly interesting period for the rise of RL in autonomous driving #ssad2026

Maximilian Igl on stage talking about Reinforcement Learning for Autonomous Driving #ssad2026
His slides are available online www.maximilianigl.com/blog/rl-for-...

Ok, I’m stealing this representation for e2e driving decoders ^^ #ssad2026

Nowadays “world models” is such loosely used. I’m super appreciative when people specify up-front what they mean by it #ssad2026

One of the inflection points for e2e driving was the switch from trajectory regression to classification over trajectories. This gives more freedom on the ways you generate trajectories but also on how to score them -> actual multimodal trajectories #ssad2026

In the last few years end-to-end driving has gone a long way from the simple L2-loss imitation and single mode prediction #ssad2026

José M. Alvarez speaks of End-to-end models for autonomous driving - from direct prediction to counterfactual deliberation on the edge #ssad2026

First session of the day by Alexandre Boulch on 3D perception in autonomous driving: from Lidar self-supervision to multi-sensor distillation #ssad2026

Let’s go #ssad2026